Coordinated processing of data by networked computing resources
By synchronizing data processing across multiple networked computing resources using timing parameters and latency monitoring, the system addresses latency issues in networked data processing systems, enhancing trade execution efficiency and reducing implicit trading costs.
Patent Information
- Application Number
- JP2021089695
- Authority / Receiving Office
- JP · JP
- Patent Type
- Patents
- Current Assignee / Owner
- Priority Date
- 2015-03-12
- Filing Date
- 2021-05-28
- Publication Date
- 2024-11-21
- Estimated Expiration
- 2036-02-26
AI Technical Summary
In networked data processing systems, orders are often routed to slower exchanges first, leading to increased latency and a lower overall fill ratio due to other parties detecting and manipulating orders, resulting in higher implicit trading costs.
A system and method for coordinating data processing by multiple networked computing resources, involving synchronization of processing requests through timing parameters and latency monitoring to ensure synchronized execution across multiple processors.
This approach reduces latency arbitrage opportunities, increases trade win ratios, and enhances fill rates by ensuring simultaneous or synchronized execution of data processing segments across networked computing resources.
Smart Images

Figure 0007588034000001 
Figure 0007588034000002 
Figure 0007588034000003
Abstract
Description
[Technical field]
[0001] CROSS-REFERENCE TO RELATED APPLICATIONS This application claims the benefit of, including priority to, U.S. Provisional Patent Application No. 62 / 126,106, entitled "COORDINATED PROCESSING OF DATA BY NETWORKED COMPUTING RESOURCES," filed February 27, 2015, U.S. Provisional Patent Application No. 62 / 126,120, entitled "COORDINATED PROCESSING OF DATA BY NETWORKED COMPUTING RESOURCES," filed February 27, 2015, and U.S. Provisional Patent Application No. 62 / 132,063, entitled "COORDINATED PROCESSING OF DATA BY NETWORKED COMPUTING RESOURCES," filed March 12, 2015.
[0002] All of the above applications are incorporated herein by reference in their entirety.
[0003] The present disclosure relates generally to systems, methods, apparatus, and computer-readable media for managing data processing by multiple networked computing resources. In particular, the present disclosure relates to the coordination or synchronization of related or temporal requests for data processing on distributed network resources.
[0004] Aspects of the subject matter disclosed herein may be underlying or related to the holding, transfer, and / or management of securities and other financial interests. Aspects of such holding, transfer, and / or management may be subject to regulation by government and other agencies. The disclosure herein is made solely in terms of logical, programmable, and communicable capabilities, without regard to statutory, regulatory, or other legal considerations. Nothing herein is intended, or should be taken or interpreted as a statement or representation that any system, method, or process proposed or described herein, or the use thereof, will or will not comply with any statute, law, regulation, or other legal requirement in any jurisdiction. Summary of the Invention [Problem to be solved by the invention]
[0005] In various forms of networked or otherwise distributed data processing systems, complex and / or multi-involved processes are often routed to multiple computing resources for execution. For example, in financial and other trading systems, orders for buying, selling, and other transactions of financial rights are often routed to multiple market or exchange servers for fulfillment. For example, when a large number of orders are routed to multiple exchanges (e.g., based on the liquidity available in each market), the orders tend to reach faster exchanges (i.e., exchanges with shorter inherent latencies) before they reach slower exchanges (i.e., exchanges with longer inherent latencies), and therefore appear on the books of different exchanges at different times. When an order begins to appear on the book of a faster exchange, other parties may detect the order and attempt to take advantage of the latency at the slower exchanges by canceling, modifying, and / or otherwise manipulating quotes (e.g., bids and offers) or other market parameters on the slower exchanges, effectively increasing the implicit cost of trading. As a result, an order that might otherwise be executed on any single exchange with a high fill ratio tends to exhibit a lower overall fill ratio when routed to multiple exchanges as split trades.
[0006] Prior art documents such as Rony Kay's paper "Pragmatic Network Latency Engineering, Fundamental Facts and Analysis" have attempted to address such problems by proposing the elimination of one-way communication (i.e., "packet") latencies. Such systems fail to address arbitrage opportunities and other problems that arise or are caused by the differences in the time it takes multiple processors to execute individual portions of multiple processors' execution requests (i.e., execution latencies) in addition to (or as part of) communication latencies. [Means for solving the problem]
[0007] In various aspects, the present disclosure provides systems, methods, and computer-executable instruction mechanisms (e.g., non-transitory machine-readable programming structures), such as software-coded instruction sets and data, for management of data processing by multiple networked computing resources. In particular, for example, the present disclosure provides systems, methods, and media useful for controlling synchronization or coordination of related requests for data processing using distributed network resources.
[0008] For example, in one aspect, the present disclosure provides a system, method, and medium for coordinating the processing of data by multiple networked computing resources. Such a system may include at least one processor configured to receive, e.g., from one or more data sources, signals representing instructions for execution of at least one data process executable by a plurality of networked computing resources; divide the at least one data process into a plurality of data processing segments each to be routed to a different one of the plurality of networked execution processors; determine a plurality of timing parameters based at least in part on latency in execution of a previous data processing request routed by the system to each of the plurality of networked execution processors, each of the plurality of timing parameters to be associated with a corresponding one of the plurality of data processing segments, the plurality of timing parameters being determined to result in synchronized execution of the plurality of data processing segments by the plurality of networked execution processors; route the plurality of data processing segments to the plurality of corresponding networked execution processors in a timing sequence based on the timing parameters; determine a win ratio for the data processing segments; and adjust the timing parameter associated with each of the plurality of networked execution processors based on the win ratio.
[0009] In certain aspects, the present disclosure provides systems, methods, and programming or other machine-interpretable instructions for effecting synchronized / coordinated processing of data by multiple networked computing resources, such systems comprising, for example, at least one processor configured to execute machine-interpretable instructions that cause the system to monitor execution of signal processing execution requests by each of a plurality of networked computing resources, determine at least one timing parameter related to latency in the execution of signal processing between the system and each of the plurality of networked computing resources, and store the at least one timing parameter in machine-readable memory accessible by the at least one processor.
[0010] Monitoring the execution of signal processing performance requests according to these and other embodiments of the present invention may be implemented continuously, periodically, and / or on any other suitable or desirable basis.
[0011] In various embodiments of various aspects of the present invention, the networked computing resources may include one or more exchange servers. The data sources may include one or more broker or trader systems or servers, the controlled signal processing may represent trades in financial rights, and the execution of the signal processing execution requests may represent execution of trades in financial rights, including, for example, stocks, bonds, options, and contract rights, currencies and / or other intangible rights, and / or commodities. In such embodiments, the requests for execution of data processing procedures may be based in whole or in part on parameters including, for example, any one or more of current market data quotes, order routing rules, order characteristics, displayed liquidity of each networked computing resource, and possible delays or latency in execution of a certain order quantity at each networked computing resource.
[0012] In the same and further aspects, the invention provides a system for controlling or otherwise managing requests for data processing by distributed computer resources, such system including one or more processors configured to execute instructions to cause the system to monitor execution of signal processing execution requests by each of a plurality of networked computing resources, determine at least one timing parameter related to latency in the execution of signal processing between the system and each of the plurality of networked computing resources, and store the at least one timing parameter for each of the plurality of networked computing resources.
[0013] According to one aspect, a system for coordinating data processing by multiple networked computing resources is provided, the system including at least one processor configured to monitor data associated with the multiple networked computing resources, including data associated with data processing segments previously routed to the multiple networked computing resources; receive signals from one or more data sources representing instructions for execution of at least one data processing executable by the multiple networked computing resources; divide the at least one data processing into at least one data processing segment to be respectively routed to one of the multiple networked computing resources based on the monitored data; determine a plurality of timing parameters, each to be associated with a corresponding one of the multiple networked computing resources and determined to result in synchronous execution of the at least one data processing segment by the multiple networked computing processors; and route the at least one data processing segment to the multiple corresponding networked computing processors in a timing sequence based on the timing parameters.
[0014] According to another aspect, a method is provided for coordinating data processing by multiple networked computing resources, the method including: monitoring data associated with a plurality of networked computing resources, including data associated with data processing segments previously routed to the plurality of networked computing resources; receiving signals from one or more data sources representing instructions for execution of at least one data processing executable by the plurality of networked computing resources; dividing the at least one data processing into at least one data processing segment to be respectively routed to one of the plurality of networked computing resources based on the monitored data; determining a plurality of timing parameters, each to be associated with a corresponding one of the plurality of networked computing resources and determined to effect synchronous execution of the at least one data processing segment by the plurality of networked computing processors; and routing the at least one data processing segment to the plurality of corresponding networked computing processors in a timing sequence based on the timing parameters.
[0015] According to another aspect, a computer-readable medium or media is provided having stored thereon instructions configured, when executed by at least one processor, to cause the at least one processor to monitor data associated with a plurality of networked computing resources, including data associated with data processing segments previously routed to the plurality of networked computing resources; receive from one or more data sources signals indicative of instructions regarding execution of at least one data process executable by the plurality of networked computing resources; based on the monitored data, divide the at least one data process into at least one data processing segment to be each routed to one of the plurality of networked computing resources; determine a plurality of timing parameters, each to be associated with a corresponding one of the plurality of networked computing resources and determined to result in synchronous execution of the at least one data processing segment by the plurality of networked computing processors; and route the at least one data processing segment to the plurality of corresponding networked computing processors in a timing sequence based on the timing parameters.
[0016] According to another aspect, a system for coordinating data processing by multiple networked computing resources is provided, the system including at least one processor configured to obtain a plurality of data processing waves, each of which identifies one or more data processing segments, one or more corresponding networked computing resources to which the one or more data processing segments should be routed, obtain a minimum handling interval for each of the networked computing resources, schedule an order for routing the plurality of data processing waves based on the wave timing sequence and the minimum handling interval for the networked computing resources, and route each of the data processing segments in the plurality of data processing waves based on the order.
[0017] According to another aspect, there is provided a method for coordinating data processing by multiple networked computing resources, the method including obtaining a plurality of data processing waves, each of which identifies one or more data processing segments, one or more corresponding networked computing resources to which the one or more data processing segments are to be routed, and a timing sequence to which the one or more data processing segments are to be routed, obtaining a minimum handling interval for each of the networked computing resources, scheduling an order for routing the plurality of data processing waves based on the wave timing sequence and the minimum handling interval for the networked computing resources, and routing each of the data processing segments in the plurality of data processing waves based on the order.
[0018] According to another aspect, a computer-readable medium or media is provided having stored thereon instructions configured, when executed by at least one processor, to cause the at least one processor to obtain a plurality of data processing waves each identifying one or more data processing segments, one or more corresponding networked computing resources to which the one or more data processing segments are to be routed, and a timing sequence to which the one or more data processing segments are to be routed; obtain a minimum handling interval for each of the networked computing resources; schedule an order for routing the plurality of data processing waves based on the wave timing sequence and the minimum handling interval for the networked computing resources; and route each of the data processing segments in the plurality of data processing waves based on the order.
[0019] According to another aspect, a system for coordinating data processing by multiple networked computing resources includes at least one processor, the at least one processor receiving, from one or more data sources, signals representing instructions for execution of a plurality of data processing operations executable by the plurality of networked computing resources, the data processing operations representing proposed trades of financial rights; obtaining data relating to available liquidity for the financial rights at each of the plurality of networked computing resources; dividing each of the plurality of data processing operations into a plurality of data processing segments, each data processing segment being split off from a single data processing to be routed to at least one of the plurality of networked computing processors; and routing each of the plurality of networked computing processors by the system. The system is configured to: determine a plurality of timing parameters to be associated with a corresponding one of the plurality of data processing segments based at least in part on latency in execution of a previous data processing request routed to the corresponding one of the plurality of networked computing processors and the available liquidity at each of the plurality of networked computing processors, the timing parameters being determined to effect synchronous execution of the plurality of data processing segments by the plurality of networked computing processors; determine a timing sequence for routing the data processing segments for all of the plurality of data processes based on the timing parameters associated with each of the plurality of data processes and the networked computing processors; and route the plurality of data processing segments to the plurality of corresponding networked computing processors based on the timing sequence.
[0020] According to another aspect, a method for coordinating data processing by multiple networked computing resources includes receiving, from one or more data sources, signals representing instructions for execution of a plurality of data processing operations executable by the plurality of networked computing resources, the data processing operations representing proposed trades of financial rights; obtaining data relating to available liquidity for the financial rights at each of the plurality of networked computing resources; dividing each of the plurality of data processing operations into a plurality of data processing segments, each data processing segment being divided from a single data processing to be routed to at least one of the plurality of networked computing processors; and determining whether or not a single data processing request has been routed by the system to each of the plurality of networked computing processors. The method includes: determining a plurality of timing parameters to be associated with a corresponding one of a plurality of data processing segments based at least in part on latency in the network and available liquidity at each of the plurality of networked computing processors, and determined to result in synchronous execution of the plurality of data processing segments by the plurality of networked computing processors; determining a timing sequence for routing the data processing segments for all of the plurality of data processes based on the timing parameters associated with each of the plurality of data processes and the networked computing processors; and routing the plurality of data processing segments to the plurality of corresponding networked computing processors based on the timing sequence.
[0021] According to another aspect, a method, when executed by at least one processor, causes the at least one processor to receive, from one or more data sources, signals representing instructions for execution of a plurality of data processes executable by a plurality of networked computing resources, the data processes representing proposed trades of financial rights and obtaining data relating to available liquidity for the financial rights at each of the plurality of networked computing resources, dividing each of the plurality of data processes into a plurality of data processing segments, each data processing segment being divided from a single data process to be routed to at least one of the plurality of networked computing processors, reducing latency in execution of previous data processing requests routed by the system to each of the plurality of networked computing processors and reducing the number of data processing requests routed by the system to each of the plurality of networked computing resources. A computer-readable medium or media having stored thereon instructions configured to: determine a plurality of timing parameters, each to be associated with a corresponding one of a plurality of data processing segments based at least in part on available liquidity at each of the networked computing processors and determined to effect synchronous execution of the plurality of data processing segments by the plurality of networked computing processors; determine a timing sequence for routing the data processing segments for all of the plurality of data processes based on the timing parameters associated with each of the plurality of data processes and the networked computing processor; and route the plurality of data processing segments to the plurality of corresponding networked computing processors based on the timing sequence.
[0022] As will be appreciated by those skilled in the art and once they are familiar with this disclosure, coordination or synchronization of the execution of distributed data processing requests, for example by synchronous or coordinated transmission of requests for such processing, has a great many possible applications in a large number of data processing fields.
[0023] Reference is now made to the drawings which illustrate, by way of example, embodiments of the present disclosure. [Brief description of the drawings]
[0024] [Figure 1A] FIG. 1 illustrates an example of a system suitable for providing data processing by multiple networked computing resources in accordance with various aspects of the present invention. [Figure 1B] FIG. 1 illustrates an example of a system suitable for providing data processing by multiple networked computing resources in accordance with various aspects of the present invention. [Figure 1C] FIG. 1 illustrates an example of a system suitable for providing data processing by multiple networked computing resources in accordance with various aspects of the present invention. [Diagram 2] 2 is a flow chart illustrating an example method for coordinating data processing by multiple networked computing resources in accordance with various aspects of the present invention. [Diagram 3] FIG. 1 illustrates an example of a system suitable for providing data processing by multiple networked computing resources in accordance with various aspects of the present invention. [Figure 4] 2 is a flow chart illustrating an example method for coordinating data processing by multiple networked computing resources in accordance with various aspects of the present invention. [Diagram 5] FIG. 2 illustrates an exemplary histogram that may be used in an exemplary method for managing data processing by multiple networked computing resources in accordance with various aspects of the present invention. [Figure 6A] FIG. 1 illustrates a comparison of performance rates when using an exemplary method and system for data processing by multiple networked computing resources versus when using conventional methods and systems. [Figure 6B]FIG. 1 illustrates a comparison of performance rates when using an exemplary method and system for data processing by multiple networked computing resources versus when using conventional methods and systems. [Figure 7] FIG. 1 illustrates the use of an exemplary metric to compare an exemplary method and system for data processing by multiple networked computing resources with results using a prior art method and system. [Figure 8] 2 is a flow chart illustrating an example method for coordinating data processing by multiple networked computing resources in accordance with various aspects of the present invention. [Figure 9] FIG. 1 illustrates an example of a system suitable for providing data processing by multiple networked computing resources in accordance with various aspects of the present invention. [Figure 10] 1 is a table illustrating exemplary data relating to networked computing resources. [Figure 11A] FIG. 2 illustrates an example schedule for routing data processing segments in a data processing wave. [Figure 11B] FIG. 2 illustrates an example schedule for routing data processing segments in a data processing wave. [Figure 11C] FIG. 2 illustrates an example schedule for routing data processing segments in a data processing wave. [Figure 11D] FIG. 2 illustrates an example schedule for routing data processing segments in a data processing wave. [Figure 11E] FIG. 2 illustrates an example schedule for routing data processing segments in a data processing wave. [Figure 12] 2 is a flow chart illustrating an example method for coordinating data processing by multiple networked computing resources in accordance with various aspects of the present invention. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0025] Throughout the accompanying drawings, like features are identified by like reference numerals.
[0026] In this disclosure, as will be understood by one of ordinary skill in the art, "synchronized" or "coordinated" may mean following any desired timing sequence, whether regular, irregular, and / or wholly or partially simultaneous.
[0027] FIG. 1 illustrates an example of a system 100 suitable for coordinating data processing by multiple networked computing resources in accordance with the present invention.
[0028] In the illustrated example, system 100 includes one or more signal sources or data sources 102 (including one or more of sources 102a, 102b, respectively), executing router processor(s) 104, and one or more networked computing resources or execution processors 106. In some embodiments, data sources 102 may include one or more internal data sources 102a that may communicate directly with router 104 (e.g., through a private local or wide area network(s) or other secure wireless or wired communication, through direct communication channel(s), or through communication within a single server(s). In the same and / or other embodiments, the data source(s) 102 may also include one or more external data sources 102b that may communicate with, for example, the router processor(s) 104 over one or more public networks 108 (e.g., a public or private telecommunications network such as the Internet) using appropriate or desired network security devices that may include, for example, data encryption, etc. In the illustrated example, the router processor(s) 104 communicates with each of the one or more networked execution or computing resources 106 over a network 110 that may be the same or different from the network(s) 108.
[0029] In various embodiments, the data source(s) 102 may include devices that provide signals to communicate data and / or instructions related to the execution of data processing processes on behalf of one or more entities generating trading and / or other data processing requests to the router processor(s) 104, which may process (e.g., aggregate by summing, averaging, etc., and / or divide into segments, etc.) and use as the basis for requests for data processing by the networked computing resources 106. The data sources 102a, 102b may include, for example, systems, servers, processors, and / or any other suitable source(s) of requests for the performance of data processing tasks, such as offers and / or bids for the purchase of instruments, intangible financial rights, etc., and / or other data processing tasks, such as language, image, and / or other communication or document processing tasks. Each or any of the data source(s) 102, processor(s) 104, and resources 106 may include multiple such systems, servers, or processors.
[0030] In various embodiments, some or all of the data source(s) 102 and router processor(s) 104 may be combined and / or otherwise configured to implement multiple programming or other machine instruction applications executing on a single machine.
[0031] The networked computing resources 106 may include any device or other resource in communication with the router processor(s) 104 to receive and execute any of a wide variety of data processing requests. Such networked computing resources 106 may include systems, servers, processors, or any other suitable devices adapted to perform any process suitable for use in implementing the present invention, including, for example, processing offers or bids for the purchase of instruments, financial rights, etc., and / or other data processing tasks, such as language or text processing, image, and / or other communication or documentation tasks.
[0032] In various embodiments, one or more data sources 102 transmit or otherwise provide signals representing instructions or requests for the performance of data processing functions to the router processor(s) 104. Instructions from any given data source(s) 102 may include instructions regarding signal processing to be performed by any one or more networked computing resources 106. The requested signal processing may include, for example, computational operations, data manipulation, and / or communication processing or other signal exchange, among others. In some, but not necessarily all, instances, such instructions may specifically identify the networked computing resource(s) 106 specifically intended for the performance of such processes.
[0033] The router processor(s) 104 may parse instruction signals received from one or more source(s) 102 and use such signals to prepare instructions or requests to be forwarded to a number of execution processors 106 for performance of data processing and / or other signal processing in accordance with the received instructions. Parsing such instructions may include, for example, identifying the type of processing(s) to be requested, including, for example, the volume or quantity of orders or bids for a trade, or the amount of document processing to be performed, and the type, nature, and / or identity(s) of the networked computing resource(s) 106 that are required to perform and are to be associated with a given data processing request and / or other signal processing request.
[0034] For example, to increase the efficiency of signal and / or other data processing functions, the router processor(s) 104 may parse, sort, and aggregate relatively smaller execution requests received from multiple sources 102 for processing into one or more larger requests, and further divide such aggregated request(s) into multiple smaller requests to be distributed to multiple execution processors 106, e.g., depending on the current capacity of the execution processors 106 to satisfy or complete such processed requests.
[0035] For example, multiple sets of instruction signals received from different data sources 102a, 102b may be associated with (e.g., addressed for delivery to and execution by) individual networked computing resource(s) 106, and such instructions may be aggregated into a single signal processing execution request directed to such networked computing resource(s) 106. In some examples, identification of the networked computing resource(s) 106 to which a given signal processing request should be imposed may occur after aggregation. For example, multiple instructions from different data sources 102a, 102b may be sorted or otherwise associated with a single signal or data processing, such instructions may be aggregated, and the aggregated instructions may be associated with one or more identified networked computing resource(s) 106 such that one or more signal processing requests may be appropriately made directed to the identified networked computing resource(s) 106. Such parsing, sorting, and / or identification may be performed according to predetermined rules or algorithms (e.g., based on the ongoing or current processing capacity of one or more particular networked computing resource(s) 106) and according to requirements encoded in the instructions or provided by the ordering source(s) 102, if applicable.
[0036] As a further example, a single set of instructions for processing data may be decomposed by the processor(s) 104 and distributed to multiple resources 106 for distributed execution. For example, a relatively large order for trading one or more financial rights originating from a single source 102a, 102b may need to be distributed across multiple exchange servers 106 to be fully fulfilled; in such a case, the request(s) from one or more source(s) 102 may be decomposed by the processor(s) 104 into orders suitable for execution by multiple such resources 106.
[0037] In some embodiments, the instruction sets may be received as part of various order waves from one or more brokers. These order waves may contain instruction sets that may be intended for transmission to various venues for trading of one or more financial rights.
[0038] The targeted or specifically identified networked computing resource / execution processor 106 may communicate with the router processor(s) 104 to receive the segmented signal processing execution requests and then execute them accordingly. Such signal processing execution may include, for example, performing text or image processing operations, mathematical calculations, or communication signal exchanges, among others.
[0039] As will be readily appreciated by those skilled in the art, various components of system 100 may be combined or implemented in the form of separate systems or devices. In various configurations, such combined or separate (sub)systems may be operated by the same or separate entities. As a particular example, one or more request source(s) 102 may be integrated with or otherwise associated with a respective router(s) 104.
[0040] In one embodiment, the system may provide one or more intelligent order routers 104 including one or more processors that may be configured for ordered, prioritized, scheduled, aggregated, segmented, and / or grouped routing of orders or other data processing segments / requests related to one or more financial interests.
[0041] For example, data processing segments may be grouped in waves, and routing of individual segments may be scheduled, aggregated, grouped, etc. based on monitored data associated with destination networked computing resources. In an embodiment, data processing segments may be routed such that orders arrive at, are processed by, and / or are executed by one or more networked computing resources within a time frame defined at least in part relative to the monitored data. Figure 1C provides a sample schematic diagram depicting several exemplary waves of data processing segments submitted by brokers X and Y to venues 1-5 using router processor(s) 104.
[0042] In some embodiments, multiple instances of a data processing request may be routed to multiple exchanges in different quantities and delivered in a timed sequence that may be based on timed timing parameters from the monitored data and may define a timed order in which initiation of routing of a data processing segment should occur.
[0043] In routing orders, the system may be configured to make timing parameter determinations, which may include determining timing ranges, distributions, etc. The timing parameters may be variable, adaptive, weighted, probabilistic, etc., and may be adjustable depending on various factors, such as network congestion, order load, order priority, venue characteristics, etc.
[0044] In some embodiments, the system may be configured to adapt the routing of orders (data processing segments) based on projected order flows. For example, the system may have a large number of scheduled orders to be routed at a particular time or during a particular time frame. Thus, the system may be configured to adapt the routing of one or more orders given projected loads on networking equipment, communication channels, venue systems, routing paths, intermediate network equipment, etc. The system may be further configured to distribute the routing of orders across various network links and / or across different time frames to balance the order routing load.
[0045] As described herein, the system monitors data associated with one or more networked computing resources, which data may be collected by or received from one or more components or devices in the system.
[0046] The system may also be configured to monitor network data associated with one or more networked computing resources, including, among other things, monitoring network performance, monitoring available network links. Probabilistic distributions and / or models of network performance may be generated, adapted, defined, and / or used in connection with the monitoring of the network. The monitoring of the network may be used to adapt time parameters, etc. The monitoring of the network may include, for example, determining and / or monitoring latency mean, maximum, standard deviation, kurtosis, L-kurtosis, skewing, median, variance, mode, correlation, cross-correlation, covariance, etc., which may be correlated and / or associated with other factors, such as order load, time of day (e.g., start of trading, end of trading), occurrence of financial events (e.g., stop trade orders, corporate events, statistical releases, analyst report releases, credit rating changes), occurrence of network events (e.g., sunspots, DoS attacks), etc.
[0047] Network monitoring may be performed, for example, through tracking network performance through "heartbeats" (e.g., transmissions of ping signals, regularly scheduled transmissions, echo request packets) to measure signal transmission times, historical order routing performance, recent order routing flows, sending test messages, etc.
[0048] In one embodiment, the test messages are used to monitor the network, and timing differences between the test messages and the routed order messages may be used in determining various order routing characteristics, such as the time it takes to process an order, internal latency within the venue, etc.
[0049] In some embodiments, the system uses various network forecasting techniques in determining order loads across various network links at various venues, etc., including orders scheduled to be routed by the system itself, and fulfillment requests may be reordered and / or scheduled based in part on the forecasted loads.
[0050] In some embodiments, the system may have access to a database(s) or may receive or monitor network messages to obtain information regarding the topology and redundancy schemes of one or more network connection(s). For example, the system may have two or more connections and / or routes to a venue. By accessing and monitoring the performance of these connections and / or routes, when there is a failure in the primary connection / route, upon detection, the system may be configured to route all subsequent orders based on the latency associated with a known secondary route / connection in order to minimize undesirable timing and possible fill rates caused by the failure.
[0051] In some embodiments, when available, monitoring of the network can be done for previously routed trade requests.
[0052] The system may also be configured to track a trade capture ratio, which may be determined by the ratio of the sum of the liquidity of the orders as determined at a certain point in the trade (e.g., when the trade decision is made) to the amount of liquidity captured.
[0053] The trade win ratio may be used for a variety of purposes, such as to adapt timing parameters, to determine that a network component has failed, to detect that a third party may be intercepting orders, etc. For example, a high trade win ratio may indicate reduced information leakage and / or increased execution rates, and a low trade win ratio may indicate increased information leakage and / or decreased execution rates. A low trade win ratio may trigger an alert, notification, or adaptation by the system through, for example, application of business logic, implementation of timing parameters, modified routing strategies, etc.
[0054] In some embodiments, the system may include business logic that may be used in determining how orders, routing instructions, etc. may be sequenced, prioritized, scheduled, aggregated, segmented, and / or grouped. The business logic may be used for route selection, route determination, route prioritization, etc. The business logic may be used to assist in making decisions. In some embodiments, the business logic may be used to prioritize orders based on various business rules, such as rules configured to prioritize venues that have order books that include a greater amount of liquidity for a particular financial security, etc.
[0055] In some embodiments, the business logic may include intelligent decision support functions, such as the ability to traverse decision trees.
[0056] In some examples, the system may access, receive, or determine information about different venues (based on monitored network and order performance). In some examples, this information may include an estimate of the message handling capacity of the venue(s). As illustrated herein, if an order queues at a venue or overloads the processing capacity of the venue(s), the resulting processing delays may affect the effectiveness of the timing parameters and may reduce the control the system has over limiting arbitrage on waiting times. Thus, in some examples, the system may be configured to use the timing parameters to avoid queuing or overloading the venue.
[0057] In some examples, venue information may include the physical location of the venues, the distance between venues (optical cable distance or transmission time), known or potentially co-located predatory traders, availability or speed of the co-located traders, etc. In some examples, one or more of these factors may be used to determine the timing parameters. For example, if the order wave pertains to a venue with known and available co-located traders, the timing parameters may be determined based on a lower timing tolerance / difference. In some examples, the routing of trades with the highest timing tolerance will occur when the trades are routed in a timing sequence that results in simultaneous arrival or execution. In some cases, simultaneous arrival / execution may not be optimal for changing the market, and thus the processor(s) may be configured to consider different tradeoffs.
[0058] The system may be configured to take into account the possibility that a third party may affect the trading and / or execution of a trade in a financial security. For example, a high frequency trader may use information related to a first order at a first venue to determine that a second order will be received at a second venue. The high frequency trader may then potentially affect the price and / or quantity available for the second order by making opportunistic trades prior to the arrival of the second order. As a result, the second order may not be filled or may be filled at a lower quantity, thereby affecting the acquisition and execution rate of the trade. As a second example, a high frequency trader may use information related to a first order at a first venue to cancel an order at a second venue and replace it with a higher priced order in anticipation of the arrival of a second order by the entity placing the first order. As a third example, a high frequency trader may use information related to a first order at a first venue to determine order information (e.g., US best offer, etc.) that may be more current than order information available at one or more venues. High frequency traders may then use this information to place opportunistic orders that result in execution of trades where the price and / or quantity is less than optimal for the counterparty of those orders, e.g., mispriced orders, etc., may remain on the order book.
[0059] The system may be implemented in various components of a trading network, for example, the system may be implemented as part of a broker electronic system, as part of a network, as an intermediate gateway and message delivery service, as part of a venue electronic system, etc.
[0060] For example, if the system is implemented as an intermediary gateway and messaging service, the system may be configured to receive multiple orders, which may be associated with one or more order waves from one or more brokers to one or more venues. The intermediary gateway and messaging service may then coordinate the routing, clustering, and / or segmentation of the orders and / or sub-orders to one or more venues. For example, a broker may "give up" an order and route the order in the broker's interest.
[0061] In some embodiments, the system is implemented at the broker level. For example, a broker may use the system for routing client orders. In some embodiments, the system is implemented at the client level. For example, a client may use the system for routing its own orders.
[0062] The system may be configured to route large numbers of orders and may be designed to scale based on order volume. Thus, the system may be implemented using a variety of suitable technologies. In some embodiments, the system is a software-based solution. In some embodiments, the system is an appliance-based solution. In some embodiments, the system is a combination of software-based and appliance-based solutions. Scalability may be important, as the system may need to be able to handle a large number of waves from multiple brokers simultaneously.
[0063] In some embodiments, the system is implemented using distributed networking technologies, e.g., cloud computing technologies. Potential advantages of using distributed networking technologies include the ability to provide resources to support various instances of the system based on order routing volume. Resiliency and / or high availability techniques, such as failover systems, managed backups, hot / cold backup schemes, etc., may be used to achieve a consistent level of service and / or uptime. For example, there may be various instances of the system that may be configured to share information with each other, which may enable lossless re-establishment of sessions to clients and venues in the event of a failure.
[0064] Certain embodiments of the system may provide various advantages, such as increased trade win ratios, increased execution rates, reduced leakage to third parties, reduced risk of "market movement," improved decision making (e.g., achieving a more optimal trade-off between order liquidity and risk of leakage), the ability to use timing information from a variety of sources, the ability to use adaptive and / or probabilistic latency models, and the ability to quickly and effectively determine faults in the network or on network equipment.
[0065] If the system is implemented as an intermediate gateway and message delivery service, there may be potential benefits since there is no / reduced integration required at the venue and / or broker level, brokers may be able to use the system with reduced need for infrastructure investment, and the system may be able to operate independently of brokers, financial institutions, clients, and / or venues. In some embodiments, existing infrastructure, messaging protocols, networks, etc. may be used in conjunction with the system. The reduced need for integration and / or adaptation, and ease of interoperability with existing standards and / or protocols may be commercially valuable in terms of regulatory requirements, resiliency requirements, security requirements, and the cost and / or complexity that may be required for system modifications. The ability to interoperate with existing external systems and / or protocols may lead to increased adaptation and / or broker flows.
[0066] In addition to or as an alternative to determining timing parameters for routing requests to multiple venues, in some embodiments, an aspect or aspects of the system may be configured to determine timing parameters for routing multiple requests to the same venue (i.e., in waves or otherwise).
[0067] An example application of the system 100 for distributed execution of segmented processing requests according to the present invention is provided by a financial system 1000 adapted for processing requests for data processing representing trades and / or offers for trades, or other transactions, of tangible and / or intangible financial rights, such as stocks, bonds, currencies (e.g., foreign exchange), various forms of natural resources or commodities, options, loans, etc. As shown in FIG. 1A and FIG. 1B, for example, in a financial transaction data processing system 1000 according to the present invention, the signal source or data source(s) 102 may include trader system(s) 1102, which may include, for example, trader / broker systems or servers as well as any other source of bids, offers, or other transactions of financial rights, such as currently provided by known financial trading platforms. In various embodiments, such trader systems 1102 may be referred to as order entry systems.
[0068] The order placement systems 1102, 102a may include, for example, systems operated by or on behalf of entities owned or controlled by a parent or other controlling organization such as a bank or brokerage house. The order placement systems 1102, 102b may include, for example, systems operated by or on behalf of a broker or other trading entity that trades through or with the assistance of independently controlled banks, institutions, and / or other brokerage houses that act on behalf of individual investors.
[0069] The router processor(s) 104, in such embodiments, may include, for example, a server(s) or other system(s) 1104 that communicates with the trader systems 1102, 102 through receipt and transmission of encoded electronic signals representing requests for data processing representing execution and / or confirmation of trades of financial interests, and communicates with a broker, exchange, or other market system or execution processor(s) 1106 for execution of such trades. In such embodiments, the processor 104 may be referred to as a Smart Order Router or a Tactical Hybrid Order Router (in either case, an "SOR") 1104, 104. The SOR 1104 may include one or more gateway(s) 1122 and / or router(s) 1124 to facilitate direct (e.g., through wired communication, using one or more dedicated communication channel(s), or through communication within a single server) and / or indirect (e.g., through wireless communication, through a network 108, 1108, or through an intermediate server) communication with one or more trader systems 1102, 102 by way of the router(s) 1104. The exchange or market system 1106, or other execution processor(s) 106, may communicate with the SOR(s) 1104 through a network 110, 1110, such as the Internet or other public network, which may be the same as the network 1108.
[0070] For embodiments of the system 100 configured as a financial trading or order execution system 1000, the requested and executed signal processing provided by the source(s) 102 may represent trades or other transactions in financial rights. Such transactions may include, for example, trades and / or offers for trades or other transactions in financial rights, such as stocks, bonds, currencies (e.g., foreign exchange), various forms of natural resources or commodities, options, loans, etc., and the networked computing resources 106 may be, for example, an exchange server 1106, examples of which may include automated or electronic market systems.
[0071] As will be appreciated by those skilled in the art, a SOR (sub)system or processor 1104 receiving such a set of trade request signals can apply a variety of processing to the request or requests. For example, if the set of signals represents a request for trading in financial rights, the requested trades can be aggregated over time and / or across multiple trade request sources 1102, and / or processing requests for trading in one or more rights can be split up for routing individually or in batches to multiple execution handlers or processors 1106.
[0072] The set of signals representing a request for trading financial interests may be, for example, a FIX session as received from a client broker. By way of example, two order types or order instructions that may be accepted include IOC (immediate or cancel) orders and directed single orders.
[0073] In conjunction with IOC orders, a broker may also send a set of signals representing orders that the broker intends to send out as a single wave to a set of exchanges. If an order is always IOC, many other functions are simplified since there may be no need to handle subsequent cancellations or replacements. Thus, IOC waves may not need to maintain affinity and may be free to use available exchange sessions to best manage congestion.
[0074] In the context of directed single orders, broker clients can use the system to transmit directed orders to exchanges utilizing network connectivity and exchange sessions. This service can be used by brokers as a primary order gateway to exchanges or as a backup service (reducing their need to back up connectivity to individual exchanges).
[0075] In some examples, other order types such as same day orders, GTC (good till cancelled) orders, or any other order type may be used, but may require additional tracking to maintain order affinity and results, including sending and receiving multiple messages between different venues (e.g., order replacement / modification / cancellation and acknowledgments).
[0076] In some embodiments, the use of Directed Orders or IOC Orders may, in some cases, enable the SOR system to employ a more efficient messaging process that may reduce order messaging load and / or processing and / or increase messaging throughput.
[0077] In various embodiments, the order source(s) 102, 1102 may be implemented together with or as part of the order router(s) 104, 1104, as described herein. For example, those skilled in the art will readily appreciate that any or all of the various components of the system(s) 100, 1000, including any or all of the processor(s) 102, 104, 106, and their methods of operation as disclosed herein, may be implemented using any device, software, and / or firmware configured for the purposes disclosed herein. A variety of components, both hardware and software, as well as firmware, when used alone and / or in various combinations, are currently known to be suitable for implementing such systems, devices, and methods, and others will likely be developed in the future.
[0078] Examples of components suitable for use in implementing systems 100, 1000 and various example processes disclosed herein, including, for example, process 200 of Figure 2 and process 300 of Figure 4, include server class systems such as, for example, IBM x3850 M2™, HP ProLiant DL380 G5™, HP ProLiant DL585™, and HP ProLiant DL585 G1™, etc. In some embodiments, a variety of other processors would be useful, including desktop, laptop, or palm model systems.
[0079] An example of a method 200 for processing a set of transaction request signals generated by a transaction request signal source 102, 1102 suitable for implementation by a router processor(s) 104, such as, for example, the SOR 1104 of the system 1000, is shown in FIG.
[0080] 2 may be considered to begin at 202 with the receipt by processor(s) 104, 1104 of signals representing requests for data processing, such as, for example, trading one or more financial interests. In an embodiment of a system 100, 1000 comprising a SOR routing processor(s) 1104 adapted to process signals representing requests for trades in financial interests and / or execution of other transactions received from a trading signal source(s) 1102, a set of signals representing requests for execution of transactions in one or more financial interests may be received from, for example, the source(s) of the request, such as a URL or other network address or identifier used by or associated with the trading system 102, 1102; The interest or interests to be traded or otherwise traded, such as an identifier used by one or more exchanges to identify stocks, CUSIP numbers for bonds, currency pairs to be exchanged, etc.; The type of transaction to be executed or requested (e.g. buy, sell, bid, offer, etc.); One or more quantities (i.e., amounts or volumes) of the right(s) to be traded (e.g., including any total quantity and / or reserve quantity); and Corresponding price terms, The signal or set of signals may include one or more identifiers representing.
[0081] The further parameters may include, for example, the following parameters, current and / or historical: The probability of completion of a multi-part or segmented trade request (i.e., the historical percentage of multi-part orders that result in a completed trade); For example, the amount of the spread between the bid and offer prices (e.g., relative to the current and / or historical trend of the spread); market volatility in the particular right being traded or in the related or corresponding right(s) or in any related benchmark or index; The depth of the market book(s), e.g., current depth against historical trend depth; Reserve quantity, -Order priority, -Tolerance for information leakage, Maximum latency (which may, for example, constrain the maximum size of relevant timing parameters), The desired routing path, Your desired venue, - specific routing instructions, Display quantity, and For example, buy side and / or sell side display size and backing.
[0082] Some or all of these parameters may be collected as or based on monitored data associated with one or more networked computing resources.
[0083] In other embodiments, such a signal set may include content and / or identifiers representing images, text, or other content to be processed by one or more execution processors 104, 1104 and specific execution requests.
[0084] If an incoming client / broker order needs to be identified as part of an order wave, in one example, the processor(s) may be configured to receive a repeating group of FIX messages, where each individual order may have a tag indicating that the order is the Nth order of a wave of M total orders.
[0085] Alternatively, orders can be marked as being the first and last order of a wave. In some embodiments, affinity may need to be preserved so that replacements and cancellations are properly handled in the same venue session as the original order.
[0086] Among the many types of market systems 1106 that are suitable for various embodiments of the present invention are alternative trading systems (ATS) of the type known as "dark" exchanges or "dark pools." Typically, such exchanges do not openly display their market offerings to members of the trading public. The use of known or predicted reserve quantities can be particularly useful in such embodiments.
[0087] Thus, examples of data records that would be provided by a source 102, 1102 to request a trade of a given right under specified terms may include the following: <Source of request (102, 1102)><Type of transaction><Rights identifier><Quantity(s)><Price terms(s)>
[0088] The signal set received by the processor 104, 1104 at 202 may be stored in any volatile and / or non-volatile memory(s) as appropriate for archival and / or further processing purposes.
[0089] At 204, the trade or other data processing execution requests received at 202 may be parsed by the router processor(s) 104, 1104 to put them into any suitable or desired form for use in creating one or more sets of instruction signals to be provided to the execution processor(s) 106, 1106. Parsing of the instruction signals may include, for example, identifying the type of trade(s) or process(es) to be requested, including, for example, the volume and / or quantity of orders or bids for trading a particular right(s), whether such volume is to be bought or sold or offered for sale or purchase, the amount and / or type of document processing to be performed, and the type and nature of the networked computing resource(s) or execution processor(s) 106 that are requested to perform and thereby to be associated with such execution or processing instructions. In various embodiments, the parsed instruction set may be stored in temporary or volatile memory(s) 118, 1018 accessible by the corresponding processor(s) 104, 1104 for aggregation with other processing requests, partitioning for routing to multiple execution processors / resources 106, 1106, and / or creation and forwarding of batches or other delayed execution requests.
[0090] In generating the one or more sets of command signals, an order scheduling module may be used that may be configured to handle downstream capacity, queuing, and congestion in the network, exchange gateways, and cross-exchange engine queues.
[0091] The module may be configured to forecast the maximum message / order flow rate that each exchange session can handle and then anticipate exceeding or approaching that rate and avoid doing so before it does. The SOR router processor(s) 1104 may be configured to re-sequence or delay the sending of earlier orders in a wave if calculation of subsequent orders in the wave or in the next wave is likely to become congested or queued.
[0092] The module may be configured with a feedback mechanism (based on ACK round-trip latency monitoring) to dynamically determine downstream congestion and queuing delays since these may not be static, consistent, or predictable.
[0093] To handle these volumes, the SOR router processor(s) 1104 may be configured to access the exchange with multiple exchange order sessions and may be configured to balance the flow load across the order sessions.
[0094] In some embodiments, the instruction set may consist of FIX messages with built-in features such as repeating groups with groups of order information stored as multiple components within a single message. These messages may often be used to carry waves of baskets, where multiple instances of the same stock order may be transmitted to multiple exchanges in different quantities to represent orders for the same stock at different venues, or to represent a portfolio wanting to buy a set of 38 stocks, and routed to a SOR router processor(s) that uses timing to determine scheduling.
[0095] The instructions received at 202 may be accumulated during regular or irregular defined time intervals, such as the duration of a business day or any segment thereof, or any other desired time period(s), which may be pre-set and / or dynamically determined by the processor(s) 104, 1104. The instructions may also be processed individually as they are received. Process 200 may return to 202 if further instructions are to be or may be received prior to processing.
[0096] The trade requests / instructions may be accumulated for a defined time interval, such as the duration of a business day or any segment thereof, or for a desired period of time, which may be preset and / or dynamically determined by the processor(s) 104, 1104. If further instructions are to be or may be received, the process 200 may return to 202.
[0097] In embodiments of the invention that employ sorting / aggregation techniques in parsing or otherwise preparing orders or other processing requests, at 206 the processor(s) 104, 1104 may repeat processes 202-204 until all required or desired related or aggregatable processing request signal sets have been received from the source(s) 102, 1102. For example, as previously described, any number of data records representing orders or requests for purchases of securities identifiable by a Committee on Uniform Security Identification Procedures (CUSIP) number may be received from the data source(s) 102, 1102 and stored in a memory 118, 1018 associated with the processor(s) 104, 1104 for batch processing, e.g.: <source 1> <sell><CUSIP No.AA><10,000><price A><res.9,000><price D> <source 2> <buy><CUSIP No.BB><12,000><price C><res.1,000><price B> <source 3> <sell><CUSIP No.BB><11,000><price A><res.8,000><price D> <source 6> <sell><CUSIP No.AA><14,000><price A><res.2,000><price E> <source 4> <buy><CUSIP No.AA><18,000><price C><res.7,000><price B> <source 1> <sell><CUSIP No.BB><20,000><price A><res.3,000><price D> <source 3> <sell><CUSIP No.AA><13,000><price A><res.6,000><price D> <source 4> <buy><CUSIP No.BB><22,000><price C><res.4,000><price B> <source 5> <sell><CUSIP No.AA><21,000><price A><res.5,000><price E> <source 4> <buy><CUSIP No.BB><15,000><price C><res.7,000><price F> <source 1> <sell><CUSIP No.AA><19,000><price A><res.3,000><price D> <source 5> <buy><CUSIP No.BB><16,000><price C><res.8,000><price F> <source 6> <sell><CUSIP No.BB><17,000><price A><res.6,000><price H>
[0098] In another example scenario, the processor(s) may receive individual large execution instructions from the data source(s) 102, 1102. For example: <source 1> <buy><CUSIP No.AAA><100,000><price A>
[0099] In another example scenario, a processor(s) may receive order execution instructions in waves, as illustrated by the example of Figure 1C. As shown in the example instructions of Figure 1C, in some examples, the incoming execution instructions may also include a specific venue field to identify the networked resource or venue to target.
[0100] Upon individual receipt, or at a given periodic rate, when a given number of orders have been received at a given time, when all desired orders have been received, or when any other desired criteria have been met, the processor(s) 104, 1104, as part of the parsing or otherwise processing instructions, may sort and / or group the stored records 204 according to any one or more desired criteria, for example, by type of trade request and rights identifier. For example, in the first exemplary scenario above: <buy><CUSIP No.AA><18,000><price C><res.7,000><price G> <source 4> <sell><CUSIP No.AA><10,000><price A><res.9,000><price D> <source 1> <sell><CUSIP No.AA><14,000><price A><res.2,000><price E> <source 6> <sell><CUSIP No.AA><13,000><price A><res.6,000><price D> <source 3> <sell><CUSIP No.AA><21,000><price A><res.5,000><price E> <source 5> <sell><CUSIP No.AA><19,000><price A><res.3,000><price D> <source 1> <buy><CUSIP No.BB><15,000><price C><res.7,000><price F> <source 4> <buy><CUSIP No.BB><22,000><price C><res.4,000><price B> <source 4> <buy><CUSIP No.BB><12,000><price C><res.1,000><price B> <source 2> <buy><CUSIP No.BB><16,000><price C><res.8,000><price F> <source 5> <sell><CUSIP No.BB><20,000><price A><res.3,000><price D> <source 1> <sell><CUSIP No.BB><11,000><price A><res.8,000><price D> <source 3> <sell><CUSIP No.BB><17,000><price A><res.6,000><price H> <source 6> As shown, the various data fields in the trade request record can be reordered or reformatted as needed or desired to suit the processing needs of the routing processor(s) 104, 1104. For example, as shown, the relevance of the "source" data item may be associated with or follow different priorities to promote efficient ordering while allowing the processor(s) 104, 1104 to report trade / request fulfillment upon completion of order processing.
[0101] The process 204 may further include aggregation by the processor(s) 104, 1104 of the received and sorted trade requests into collected or consolidated order(s) for trading a particular type of particular right(s), for example, by summing total or subtotal quantities associated with the corresponding trade requests, thus: <buy><CUSIP No.AA><18,000><price C><res.7,000><price G> <sell><CUSIP No.AA><77,000><price A><res.18,000><price D> <res.7,000><price E> <buy><CUSIP No.BB><65,000><price C><res.15,000><price E> <res.5,000><price B> <sell><CUSIP No.BB><48,000><price A><res.11,000><price D> <res.6,000><price H>
[0102] 9 illustrates a process and data flow diagram of example processes and data flows for an order router / system 104, 1104. In some examples, the order router / system 104, 1104 may include a client message processing module 2105 that may configure the processor(s) 104, 1104 and / or other hardware components to receive waves of client messages. In some examples, client order instructions may be received in waves of multiple messages. In some such examples, the client message processing module may be configured to stage order messages in a buffer or other order wave staging memory / storage device 2110 until all order messages relating to a particular wave have been received.
[0103] In one example, a data record to be provided by a source 102, 1102 to request a trade of a given right under specified terms may include, in addition to other fields containing details of the trade request: <Wave ID><Total number of orders in wave><Current order index> <Timing mode for wave><Timing offset for manual timing mode><Security key(s)>
[0104] In another example, instead of including a data field indicating the total number of orders in a wave, the data record may include a <wave complete> flag to indicate that the current order message is the last message in the wave.
[0105] In some examples, the incoming order message may be in FIX (Financial Information eXchange) format, or any other standard or proprietary format / protocol.
[0106] As shown in the example of FIG. 9, as individual orders or complete waves are received, they may be stored in one or more buffers or other memory / storage until they are ready to be processed by the wave timing handler 2115. In the example system of FIG. 9, complete order requests (including individual orders and complete wave orders) may be stored in multiple levels or queues. For example, order requests may be stored initially in client session queues until they are selected to be moved to a wave queue. In some examples, the system is configured to allow only N waves to be in a wave queue. In some cases, this can prevent a high volume session from dominating the wave queue and routing resources. In some examples, the system can be configured to allow more than N waves to enter a wave queue when other session queues are not being used to their quota.
[0107] In other examples, the complete order request may be stored directly in one or more wave queues (ie, without any intermediate client session queues).
[0108] The wave timing handler 2115 may be implemented by one or more hardware device(s) and / or processor(s) 104, 1104 configured to manage the queue(s) and determine which order(s) in the queue(s) will be transmitted to the venue next.
[0109] In some examples, a modified first-in, first-out (FIFO) methodology may be used by the wave timing handler. However, rather than strictly FIFO, in some embodiments the timing handler may be configured to reorder, cherry-pick, skip, or otherwise select the order(s) in the sequence other than simply traversing the queue. For example, in some embodiments, the first M wave orders in the queue may be processed to determine a sequence for routing these M wave orders, or simply determine which of the M wave orders will be routed next, without regard to subsequent orders or otherwise.
[0110] In some examples, the wave timing handler may be configured to determine a sequence in which to route wave orders to increase out of available order routing throughput, to decrease the time it takes to route all of the orders, and / or to optimize any other aspect of the operation of the router(s) 104, 1104.
[0111] In some examples, as described herein or otherwise, the ordering of the processing of waves can be based on venue latency, throughput, market data (e.g., available liquidity, pricing, etc.), etc.
[0112] When all desired signal sets have been received at 202, and optionally sorted, accumulated, and / or otherwise processed at 204, the processor(s) 104, 1104 may use the instruction sets processed at 204 to create an execution request signal set for transmission to the resource / execution processor 106, 1106 at 208. Such execution request signal sets may include any necessary or desired signals to effect the requested processing, including content or data and command signals. For example, in an embodiment adapted to process requests for trading financial rights, the requests may be sorted and / or aggregated based on the right(s) to be traded, the quantity of the right(s) to be traded, price, etc., and associated with the appropriate execution command signal. The form of any execution command signals associated with a given request may depend on the nature and type of requests to be executed and the processor 106, 1106 that is to execute them, as well as any network 110, 1110 over which signals exchanged between the processor(s) 104, 1104 and 106, 1106 are to be transmitted, including applicable protocol and instruction format requirements, as will be appreciated by those skilled in the art. Thus, data relating to any or all of the systems 106, 1106, 104, 1104, and 110, 1110, protocols used thereby, and / or information related to rights traded, offered, or described thereby may be accessed and used by the processor(s) 104, 1104 in parsing and creating instructions for the execution of operations by any of the processors or resources 106, 1106. Sources of such data 1126 may include, for example, exchange market data system 1126v (FIG. 1B), which, in embodiments of the invention adapted for processing financial transactions, may include information received from various exchange systems 1106, news information sources such as Bloomberg or Reuters, and / or other sources.
[0113] In assembling a request for data processing using networked processing resources, including many resources configured for use in executing financial transactions, it may sometimes be necessary or desirable to break down an execution and / or other processing request into multiple parts. Such parts or segments may correspond to portions of a larger order or other data processing request to be executed by multiple networked resources 106, such as, for example, exchange servers or other execution processors or handlers 1106. For example, when multiple exchange servers or other markets are available for execution of a trade request representing a purchase order for a large amount of financial rights, such as stocks or bonds, it may be necessary or desirable to split the order into multiple parts for execution in multiple markets and / or by multiple exchange servers 1106. For example, a sufficient quantity of a particular right may not be available on a single exchange at all or at a desired price, and it may be necessary or desirable to break down a single order into smaller segments and route them to multiple exchanges in order to fulfill the order in its entirety.
[0114] Thus, for example, in various embodiments of the present invention directed to processing requests for transactions of financial instruments, when a router 104, 1104 is requested by one or more sources 106, 1106 to complete a transaction of one or more financial rights, in creating a set of signals representing the request for the transaction, the router 104, 1104 can access information available from sources such as market data source(s) 1126 as well as from any one or more execution processor(s) 106, 1106 to determine the quantity of such rights available through each processor 106, 1106 and the conditions under which such quantity is available, and can construct an execution request signal set configured for routing to each desired processor 1106, 1106 based on the quantity available under the most favorable conditions.
[0115] For example, continuing with the first example above, it may be necessary or desirable to split one or more incoming processing requests into smaller parts that are directed to multiple exchanges in order to obtain fulfillment of the complete order(s). This may be accomplished, for example, by accessing data representative of the current order book provided by one or more of the exchange servers 1106 and using known data processing techniques to split the order(s) appropriately. Thus, for example, the aggregated "sell CUSIP No. AA" order above may be broken down into portions or segments, and associating data representative of such segments with URLs or other network resource address identifiers suitable for use in routing the various segments to multiple exchange servers A1-C3 as desired, thus: <exchange A1> <sell><CUSIP No.AA><15,000><price A><res.6,000><price D><res.2,000><price E> <exchange B2> <sell><CUSIP No.AA><27,000><price A><res.6,000><price D><res.2,500><price E> <exchange C3> <sell><CUSIP No.AA><35,000><price A><res.6,000><price D><res.2,500><price E>
[0116] As described herein or otherwise, in some examples, creating 208 an execution request may include combining or splitting an order into fewer or more orders to multiple different venues. In some examples, creating 208 an execution request may include formatting, reordering, or otherwise preparing the original order, split orders, and / or combined orders into FIX or other format(s) compatible with each destination venue.
[0117] In one embodiment, in generating an execution / trade request, the SOR routing processor(s) 1104 may be configured to determine to which venue the order should be routed. Various factors may be considered by the SOR routing processor(s) 1104, such as quote size, cost structure, support for various order types, rebate structure, average execution speed, latency, latency variance, historical execution rates, etc.
[0118] 2 illustrates making 208 execution requests and determining 210 the timing for the execution requests as separate blocks, in some embodiments, making the requests and determining the timing for the requests may occur in reverse order, may occur simultaneously, and / or may be part of the same process. For example, various factors and data described herein or otherwise observed / measured may be used to both make execution requests and determine the timing for those requests.
[0119] As will be appreciated by those skilled in the art, execution of individual portions of a distributed trade, such as trading financial rights located on multiple exchanges, or other multi-part data processing request, typically requires different amounts of time by multiple network resources, such as market or exchange servers 1106 or other execution processors 106. That is, when multiple parts of a desired trade execution request are sent simultaneously to multiple exchange execution processors 106, 1106, each part or segment of the trade request may be expected to execute at a different time. This is because of the amount of time or "latency" required for the transmission of the execution request signal from the order router(s) 104, 1104 to the various different resources or execution processors 106, 1106 across the network 110, 1110 or other communication paths; the actual processing of the corresponding portions of the execution request by the corresponding processor 106, 1106; and / or the transmission of confirmation data or other data back to the order router(s) 104, 1104, which typically varies depending on several factors including, for example, the network path between the router(s) 104, 1104 and the execution processor 106, 1106; the amount of network traffic processed by the network(s) 110, 1110; the number of requests handled by an individual execution processor 106, 1106;
[0120] For several reasons, it may be important in such cases to synchronize the execution of two or more portions of a multi-part execution request. As one example, when an execution request represents a request for the execution of multiple parts of a financial transaction in multiple markets or on multiple exchanges, unsynchronized aggregate execution of individual portions of the transaction by multiple corresponding servers may affect both the likelihood of completing later portions of the transaction and / or the conditions under which such later portions may be completed.
[0121] A specific example where it is desirable to synchronize execution requests is illustrated through reference to Figure 3. In the example shown in Figure 3, the system 100, 1000 includes an order router 104, 1104 and a number of networked execution resources 106, exchange servers or execution processors 1106 "Exchange 1", "Exchange 2", "Exchange 3". In addition, the system 100, 1000 of Figure 3 further includes a co-located trading server 304 configured to execute trades or other transactions on the execution resource 1106 "Exchange 1". As illustrated in the figure, the co-located trading server 304, using a relatively low latency trading algorithm, is associated with Exchange 1 in a manner such that it can execute transactions with Exchange 1 in a relatively short amount of time compared to the amount of time it takes other processors, such as the router(s) 104, 1104, to complete similar transactions with Exchange 1. For example, the co-located server 304 can be communicatively linked to Exchange 1 by a direct wired connection or other rapid processing system. Furthermore, Exchange 1 can complete execution requests with non-co-located processor(s) 104, 1104 in a relatively shorter time (i.e., with "less latency") than either Exchange 2 or Exchange 3. In other words, as shown in Figure 3, latency X<Time Y and Time X<Time Z, while the execution time for a transaction between the co-located server 304 and Exchange 1 is less than any of Time X, Time Y, and Time Z.
[0122] For example, if a signal representing a request to trade one or more financial rights is received by the Order Router Processor 104, 1104 from one or more request sources 102, 1102, and the request is in a quantity or size such that an order reflecting the request would be too large to be fulfilled in its entirety by any one of Exchanges 1, 2, or 3, the Order Router 104, 1104 may check availability on the various available processors 106, 1106 and attempt to split the order accordingly, with portions routed to each of Exchanges 1, 2, and 3. When the Routers 104, 1104 of FIG. 3 simultaneously transmit divided portions or segments of a request for execution of a requested trade to each of the Execution Processors 106, 1106, Exchange 1, Exchange 2, and Exchange 3, the Trading Server 304 (which may be operated by, e.g., a high-frequency trading entity or other speculative investor) may fulfill a portion of the trade on Exchange 1, for example, by acting as a counter party to the proposed trade by selling or buying all or a portion of the trade request forwarded to that exchange by the Order Router 104, on terms specified in the request for the trade, and may have time to modify or otherwise advertise terms for fulfilling the remaining portions of the order on Exchange 2 and / or Exchange 3, on terms more favorable to the party making the trade(s) available (e.g., the party operating or acting through the Server 304) than terms that the offerer of such trade (e.g., the person behind the order provided by the Request Processor(s) 104, 1104) might have otherwise sought.In other words, for example, co-located trading server 304 may be able to execute a portion of a requested trade on Exchange 1 due to differences in execution latency associated with trades with Exchange 1, Exchange 2, and Exchange 3, and then move to improve its terms by raising or lowering its bid / offered price to execute the remaining portion of the trade on Exchange 2 or Exchange 3 before such remaining portion can be executed at the previously specified price, e.g., to increase its operator's or beneficiary(ies)' own profits, or the profits of other traders offering similar rights on these exchanges.
[0123] As can be seen in Figure 3, such a possibility (which can be called a "latency arbitrage" opportunity) may exist when: Time X + Time A < Time Y, and / or Time X + Time B < Time Z
[0124] Those skilled in the art will appreciate that even if a trade or other processing request signal is sent simultaneously from router(s) 104, 1104 to each of exchanges 1, 2, 3, the time it takes for each divided portion of the request to be received, acknowledged, and / or processed by the respective resource 106, 1106 (e.g., times X, Y, Z) may generally differ due to, for example, differences in network communication paths and processing speeds in any or all of the processor(s) 104, 1104 and / or 106, 1106. Similarly, the time it takes trading server 304 to change the terms of a trade it offers at each of exchanges 2 and 3 may generally differ.
[0125] Drawbacks that may arise in such cases include that traders represented by the request source(s) 102, 1102 may pay a higher price than they would otherwise have paid to execute their trades in the absence of such arbitrage opportunities, or that traders may not be able to complete trades in the desired quantities if subsequent prices on the exchange change enough to fall outside the conditions specified in their execution requests; for example, all or a portion of a trade routed to the exchange processor 1106 may not trade due to the changed price.
[0126] In instances where trade instructions may not be fully fulfilled at an exchange server 1106, e.g., due to manipulation of prices or other conditions by latency-based third parties, it may be useful to time or schedule the transmission of trade requests to multiple exchange servers 1106 in issuing data processing requests at one or more exchange servers 1106 such that execution of such trade requests at all exchange servers 1106 occurs in a synchronized manner, e.g., in a substantially simultaneous manner. In particular, it may be useful to synchronize execution of signal processing execution requests, or portions or segments thereof, at multiple networked computing resources 106, 1106, e.g., such that signal processing is received, acknowledged, and / or executed by the resources 106, 1106 in a substantially simultaneous manner.
[0127] In some examples, the signal processing to be performed in each processor 106, 1106 may not necessarily be performed simultaneously, but may be performed in a timed manner such that execution of the request(s) or a segment thereof occurs before any change in conditions can be implemented by the trading server 304. Time Y - Time X < Time A, and / or Time Z - Time X < Time B, In some cases, the use of such synchronized timing may be sufficient, e.g. Time X + Time A > Time Y, and / or Time X + Time B > Time Z , thus negating arbitrage opportunities in latency, for example. In one embodiment, the present invention therefore provides the router(s) 104, 1104 with the ability to execute trades across multiple resources 106, 1106 with minimal or no time variance, such that algorithms executed by trader(s) 304 using low latency algorithms are not given enough time to react to market changes.
[0128] In some embodiments, the timing range required to ensure that arbitrage at the latency described in the above equation can be minimized and / or eliminated can determine the acceptable range for routing of signals representing orders to one or more trading servers 304.
[0129] Thus, in these and other cases where synchronization is desired, at 210 the processor / router 104, 1104 may determine absolute or relative timing to be assigned or otherwise associated with various portions or segments of an execution request to obtain a desired sequencing. Such timing may be determined to provide any desired synchronization, for example, timing configured to provide simultaneous or substantially simultaneous execution, or timing configured to provide any desired sequencing.
[0130] In some examples, sequences for routing requests can be defined by determining acceptable timing parameters (including ranges) to stagger the routing of different routing requests. These sequences can include requests to different venues or multiple requests to the same venue. In some examples, timing sequences can be determined such that related requests are executed within a defined time threshold. This time threshold can be based on the timing of arbitrage with observed latencies or other methods.
[0131] Thus, at 210, timing parameters may be determined for each signal processing execution request or portion thereof to be assigned to each respective networked computing resource 106, 1106. The parameters are determined in a manner to effect synchronized execution of the signal processing execution requests at each of the networked computing resources 106, 1106. This determination may be based at least in part on corresponding determined latencies in the execution times of such request(s) and / or portion(s), such as, for example, any or all of the latencies A, B, X, Y, Z of FIG. 3 and / or any other applicable latencies in the execution of signal exchanges between the router processor(s) 104, 1104 and each of the networked computing resources 106, 1106 or in the processing of other such signals by any of such devices.
[0132] In some embodiments, the timing parameters may be assigned to each signal processing execution request rather than to the networked computing resources 106, 1106. In some embodiments, the timing parameters may be defined as ranges for execution and used according to business logic applied to determine and / or implement a routing strategy for association with one or more respective signal processing execution requests. For example, the timing parameters may be used to indicate that a particular set of signal processing execution requests should be processed / received / executed within a particular time range, where the time range is dynamically determined by the SOR router processor(s) 1104. For example, the time range may be determined based on various business logic flows that receive liquidity on order books of a particular venue as a parameter. In some embodiments, the time range and / or the timing parameters may also be satisfied through the selection of a particular routing path or communication link.
[0133] In some embodiments, the timing parameters may take into account monitored network conditions and / or characteristics, such as variance, standard deviation, median, mean, mode, kurtosis, and the like.
[0134] In some embodiments, the timing parameters may be determined based on timing ranges and / or probability(s) that a particular trade will arrive and / or be executed within a particular latency range. As illustrated in FIG. 5, historically monitored latency data may, in some instances, provide probability(s) and / or other statistical data such as standard deviation, variance, etc. In some embodiments, the SOR router processor(s) 1104 may be configured to determine timing parameters that fall within defined statistical thresholds.
[0135] As an alternative or in conjunction with the latency required for transmission, network characteristics may be monitored and used by the SOR router processor(s) 1104 for routing orders.
[0136] Monitoring network performance can be a critical task since network conditions can be constantly changing due to a variety of factors such as network congestion, traffic, available bandwidth, channel strength, noise, artifacts, network events, etc., and performance can affect order transmission and / or routing. There may be venue-specific network issues such as congestion at an exchange gateway, etc.
[0137] In some embodiments, the monitored network performance may be estimated and / or modeled using one or more measurements to determine statistical values such as, for example, mean, mode, median, variance, covariance, correlation, kurtosis, skew, standard deviation, etc.
[0138] The monitored network performance may be monitored at various levels of granularity and scope, for example, individual network paths, venues, etc. In some embodiments, there may be multiple network paths that may be used to route signals to venues, and the paths themselves may be monitored for their network characteristics.
[0139] For example, there may be three different network links A, B, and C, and if A and B are congested, then network link C may be selected.
[0140] In some embodiments, the network performance monitors may be weighted such that more recent network measurements have greater significance and / or influence on the model to reflect the freshness of the measured information.
[0141] Network monitoring may be based on a variety of data, for example, scheduled return trip heartbeat messages may be sent to periodically determine network latency across certain links. In some embodiments, historical order and / or routing information may be used in determining network performance, for example, router 1104 sent 12 orders on this link in the past 2 ms and encountered certain latencies with particular characteristics.
[0142] In some embodiments, test messages, duplicate messages, etc. may be transmitted to determine network latency characteristics. Messages of various lengths and / or with or without instructions may be compared to one another to determine latency associated with venue processing.
[0143] In some embodiments, the monitoring of the network may be configured to determine different components of trade latency. For example, when round trip latency is monitored, the round trip latency may be divided between two or three components. For example, the three components may include latency associated with the transmission of a request sent to the venue, latency associated with the execution of the request by the venue, and latency associated with the transmission of a response sent from the venue. In another example, the round trip latency may be divided between the transmission latency (i.e., a combined measure of the transmission latency for sending a request to the venue and the transmission latency for receiving a response from the venue) and the latency associated with the execution of the request by the venue. Other combinations of different latency related components may be used.
[0144] In some examples, the determination of different components of the round trip or other aggregate latency may be based on monitoring aggregate latencies for different request types. For example, requests for different order types or for quotes may result in different aggregate round trip latencies, which in some examples may be used to determine different components of the aggregate latency and / or to determine timing parameters for the different request types.
[0145] In some examples, the processor(s) may be configured to select and / or adjust the timing of targeting venues based on other factors as described herein or otherwise at 208 and / or 210. In some examples, the processor(s) select and / or adjust the timing of venues based on data related to the venues. This data, in some examples, may be accessible from a data source and / or may be compiled or observed based on acquisition ratios, wait times, etc.
[0146] In some examples, the processor(s) may be configured to favor or have longer timing differences for venues that are less likely to have more competition for trade requests (e.g., from co-located and / or high frequency traders). In some examples, the processor(s) may be configured to have different preferential rankings or timing adjustments for different venues based on factors that may indicate more competition or a higher risk of a trade losing to arbitrage. In some examples, explicit knowledge of opportunistic co-located traders may be accessed from a data source or may be implicitly determined based on lower winning ratios and / or lower timing differences associated with a particular venue.
[0147] In some examples, the processor(s) may be configured to have different preferential rankings or timing adjustments for venues based on the fee and / or rebate structure of the venue. For example, venues that offer rebates for posting certain trades may be more likely to have competition from high frequency traders.
[0148] In some examples, when multiple venue options are available, the processor(s) may be configured to select a venue that is less likely to have more competition. In some examples, the processor(s) may be configured to generate timings (e.g., schedule it later in the sequence or have a larger delay) that may risk missing liquidity from a venue with more competition if the opportunity to capture liquidity at other venue(s) increases. In some examples, the processor(s) may generate tighter timings for waves associated with venues with more competition.
[0149] In some examples, the processor(s) may be configured to select a venue and / or adjust the timing difference based on lost liquidity or lost gross value that would have been experienced if the trade request had arrived at the venue too early or too late, which in some examples may be based on posted liquidity, posted bid / ask pricing, etc. The processor(s) may be configured to lower the preferred ranking or narrow the timing difference for venue(s) that would have resulted in a larger loss (liquidity or value) if the trade request had arrived at the venue(s) too early or too late, potentially resulting in losing arbitrage.
[0150] In an embodiment, the SOR router processor(s) 1104 may be configured to automatically adjust the values of the timing parameters and / or the size of the timing ranges in response to various measured values provided as feedback, such as trade win ratios, recent execution rates, etc. For example, a decreasing trade win ratio may indicate information leakage, and the ranges for the timing parameters may be automatically narrowed and / or the timing parameters may be modified accordingly.
[0151] For example, in addition to latency and other parameters, the SOR processor(s) may be configured to monitor execution rates or capture ratios of previous orders routed along a particular route or to a particular venue. In some examples, the processor(s) may be configured to determine execution rates or capture ratios by comparing the total trading volume for all routed related trade requests to the available liquidity. In some examples, the available liquidity may be the total posted liquidity at each of the targeted venues. In some examples, the available liquidity may be the total posted liquidity at any venue. In some examples, the available liquidity may be based on posted liquidity data available to the SOR processor(s) when creating an execution request, which may include the processor(s) selecting venues to target and / or splitting large order requests. In some examples, the available liquidity may be based on posted liquidity data available to the SOR processor(s) at the time the first trade request in a wave or sequence is routed.
[0152] In some embodiments, the available liquidity may include publicly posted liquidity and / or forecasted liquidity. The processor(s) monitoring data associated with the networked computing resources may be configured to monitor whether publicly posted liquidity acquired by a previously routed data processing segment is subsequently replenished with new publicly posted liquidity and / or the timing between acquisition of the original liquidity and re-posting of the new liquidity. The creation of the data processing segments and the determination of their timing parameters may be based on this forecasted liquidity data.
[0153] In another example, the processor(s) may track or identify when multiple data processing segments (unrelated or related to the same data processing) target the same posted liquidity. By monitoring data related to these data processing segments to see if their performance rates exceed the posted liquidity, the processor(s) can compile this data as obtained unposted liquidity data.
[0154] Monitoring the data associated with the networked computing resources may include monitoring both re-postings of liquidity and acquisition of non-posted liquidity as predicted liquidity data that may be used by the processor to create data processing segments and their timing parameters to target such predicted liquidity data.
[0155] In one example, the processor(s) may be configured to determine a fulfillment rate or capture ratio by compiling confirmation / acknowledgement or other response messages from targeted venues to determine the total trading volume successfully captured / completed after all trade requests in a wave have been routed.
[0156] In some examples, the processor(s) may be configured to adjust the timing parameters and / or timing sequences based on a determined performance rate or win ratio. For example, if the win ratio for a route or venue drops from its previous, historical, and / or typical value, the processor(s) may determine that the timing parameters and / or sequences (e.g., based on wait times and / or other factors) are less effective at avoiding any information leakage or arbitrage trades.
[0157] In some examples, when the acquisition ratio falls below a defined threshold or by more than a certain threshold amount, the processor(s) may be configured to adjust a timing parameter (e.g., decrease or increase time Y or time Z in FIG. 3 ), adjust a timing sequence, and / or decrease a timing threshold(s) relating to the difference in execution times at different venues (e.g., decrease time=A and / or time=B in FIG. 3 ).
[0158] In some examples, when the acquisition ratio drops more than a defined threshold or by more than a certain threshold amount, this may indicate a change in the network device, such as a failure or near failure of a link or component in a router or other switching device. In some examples, the processor(s) may be configured to generate an alert or notification that there may be a possible network problem and / or that an equipment problem may exist. The alert or notification may include an audio alert, a visual alert on a display or light source, a message(s) (e.g., email, instant messaging, etc.), or any other mechanism for alerting an administrator that a network problem and / or equipment problem may exist. In some examples, it has been observed that a dropping acquisition ratio can provide an early warning of an equipment problem or a predictive failure condition before the router or other affected equipment generates its own alert or failure notification.
[0159] In some examples, the defined threshold or threshold amount may be an absolute or relative value. For example, the processor(s) may be configured to adjust the timing parameter(s), sequence(s), and / or threshold(s) when the acquisition ratio drops by 2% or drops by 2% of its previous or historical value.
[0160] In another example, the processor(s) may be configured to adjust the timing parameter(s), sequence(s), and / or threshold(s) when the acquisition ratio falls outside its typical distribution range or varies by a defined number of standard deviations from its norm (e.g., when the acquisition ratio deviates from the norm by 1-3 standard deviations).
[0161] In one example, the processor(s) may be configured to determine the variance range and / or standard deviation by collecting acquisition ratios over an X time period and creating a Gaussian or other distribution.
[0162] Upon determining that the winning ratio for a route / venue or route / venue pair or group falls below or has decreased by a defined threshold, the processor(s) may be configured to automatically adjust the timing parameter(s), sequence(s), and / or timing difference threshold(s) for the associated route(s) and / or venue(s) to attempt to increase the winning ratio for future trade request waves or sequences.
[0163] In some examples, the processor(s) may be configured to select and / or adjust the timing of targeting venues based on other factors as described herein or otherwise at 208 and / or 210. In some examples, the processor(s) select and / or adjust the timing of venues based on data related to the venues. This data, in some examples, may be accessible from a data source and / or may be compiled or observed based on acquisition ratios, wait times, etc.
[0164] In some examples, the processor(s) may be configured to favor or have longer timing differences for venues that are less likely to have more competition for trade requests (e.g., from co-located and / or high frequency traders). In some examples, the processor(s) may be configured to have different preferential rankings or timing adjustments for different venues based on factors that may indicate more competition or a higher risk of a trade losing to arbitrage. In some examples, explicit knowledge of opportunistic co-located traders may be accessed from a data source or may be implicitly determined based on lower winning ratios and / or lower timing differences associated with particular venues.
[0165] In some examples, the processor(s) may be configured to have different preferential rankings or timing adjustments for venues based on the fee and / or rebate structure of the venue. For example, venues that offer rebates for posting certain trades may be more likely to have competition from high frequency traders.
[0166] In some examples, when multiple venue options are available, the processor(s) may be configured to select a venue that is less likely to have more competition. In some examples, the processor(s) may be configured to generate timings (e.g., schedule it later in the sequence or have a larger delay) that may risk missing liquidity from a venue with more competition if the opportunity to capture liquidity at other venue(s) increases. In some examples, the processor(s) may generate tighter timings for waves associated with venues with more competition.
[0167] In some examples, the processor(s) may be configured to select a venue and / or adjust the timing difference based on lost liquidity or lost gross value that would have been experienced if the trade request had arrived at the venue too early or too late, which in some examples may be based on posted liquidity, posted bid / ask pricing, etc. The processor(s) may be configured to lower the preferred ranking or narrow the timing difference for venue(s) that would have resulted in a larger loss (liquidity or value) if the trade request had arrived at the venue(s) too early or too late, potentially resulting in losing arbitrage.
[0168] In some embodiments, the router processor 1104 may be configured to optimize order flow across multiple routes and for multiple orders. For example, increasing the fill rate for a single route through a particular routing scheme may penalize other routes. The router processor 1104 may be configured to balance order loads across venues, communication links, network links, time, etc. The router processor 1104 may be configured to determine possible latency impacts that may result from congestion caused by the scheduled routing of the router processor 1104 and schedule and / or reorder and / or associate timing parameters accordingly. In some embodiments, the router processor 1104 may be configured to automatically rearrange and / or shuffle scheduled transmissions.
[0169] In some embodiments, the router processor 1104 may be configured to apply one or more business rules that represent business logic. The business rules may effect the skewing and / or weighting of various factors used in determining the routing characteristics for a particular set of instructions associated with an order. For example, the rules may determine how an order is segmented, to which venue(s) the order may be transmitted, acceptable time ranges, etc.
[0170] For example, a particular venue having a particular latency may be favored because that venue may have a high trade win ratio associated with it and a higher weighting may be associated with routing signals to that particular venue.
[0171] As another example, a first venue may offer a thicker order book with greater liquidity than a second venue. The first venue may have somewhat less desirable network characteristics, but may be weighted more heavily for selection given this amount of liquidity available.
[0172] In an example embodiment, the SOR processor(s) may be configured to determine which venues to target by weighting liquidity and latency distributions, as well as timing parameters for those venues.
[0173] In some examples, latency variance may be caused by factors such as network traffic, poor or varying signal quality of network communications, network node problems, congestion or overload at the venue (networked execution processor(s)).
[0174] In some examples, latency variance may result from delay(s) intentionally introduced by the venue or aspects of the venue's communication network or device(s). For example, a venue may randomly introduce a delay (e.g., a "speed bump") to certain types of order requests or to all orders. This delay will increase the latency for the execution of the affected order requests. In some examples, a venue(s) may introduce a fixed delay (e.g., 350 μs) to order requests. In some examples, a venue(s) may introduce a delay having a randomly selected length (e.g., between 5-25 ms) to order requests.
[0175] In some examples, the order request to which delay(s) are applied and / or the length of the delay(s) may be randomly selected may be based on the order type, which may be based on order parameters / options / flags / etc., and / or any combination thereof.
[0176] In some instances, the venue(s) may introduce a delay having a randomly selected length selected from a range based on the order request type / parameters / options / flags / etc. For example, one classification of orders may be subject to a random delay between 1-10 ms, while another classification or order may be subject to a random delay between 14-24 ms.
[0177] Data associated with previously routed data processing segments, which may have experienced such random delays monitored by the processor(s), may in some cases exhibit a multi-modal distribution, for example in the range of 1-10 ms and in the range of 14-24 ms.
[0178] In some examples, the processor(s) may be configured to access defined aspects of the intentionally applied delay from one or more database(s) or other data source(s). In some examples, the intentionally applied delay aspects may be observed by tracking historical orders and their respective processing times. Based at least in part on the accessed and / or observed delay information, in some examples, the processor(s) may be configured to determine latency ranges, variances, and other statistics related to latency.
[0179] Data relating to previously routed data processing segments, which may be subject to randomly introduced delays monitored by the processor(s), may in some cases exhibit a multi-modal distribution exhibiting a range of latencies that does not include random delays and a second range of latencies that does include random delays.
[0180] In one example, the processor(s) may be configured to obtain data related to the available liquidity for a proposed trade for each networked computing resource (venue) that may be used.
[0181] In some examples, data related to available liquidity may be received from publicly available sources and may include the number of shares openly available at the venue. In some examples, the processor(s) may be configured to determine whether more liquidity may exist than is openly posted (e.g., iceberg or reserve volume). This determination may be based on data obtained regarding historical or past orders received at the venue for similar financial rights, volume, time, counterparty, and / or any other order data.
[0182] At 208 and 210, the processor(s) may be configured to split the large proposed trade into smaller data processes representing smaller trade requests to different venues or other networked execution processors based on the latency and liquidity data. In some examples, the processor(s) may be configured to select the venue(s) and volume based on the liquidity data and the latency distribution.
[0183] In one example, if a venue has a large amount of available liquidity but a large latency variance, the processor(s) may be configured to balance the opportunity to acquire that liquidity with the opportunity to not acquire liquidity at other venues due to market movements over a range of possible latency periods.
[0184] For example, in a hypothetical situation, venue A has 5000 shares available at price X and has a latency of 100 μs+ / -80 μs, and venue B has 500 shares available at price X and has a latency of 80 μs+ / -1 μs. If the SOR is trying to acquire all 5500 shares, the processor(s) may be configured to route a first trade request to venue A at t=0 and a second trade request to venue B 20 μs later. If the trade requests arrive or are processed according to the average latency, the trade should acquire the volume available at both venues at t=100 μs. However, because venue A's latency variance is so large, the trade at venue A may be able to execute at t=20 μs, and the price or availability of the 500 shares at venue B may have changed before the trade can be executed at venue B at t=100 μs. Similarly, if a trade at venue A does not execute until t=180 μs, an order at B that executes at t=100 μs may trigger the market to modify or cancel the order at A before the request can be fulfilled at t=180 μs.
[0185] In some examples, the processor(s) may be configured to select the networked computing resource(s) / venue(s) based on the ratio of liquidity to latency variance. When available liquidity at a venue is high, the processor(s) may be configured to target that liquidity even if it has a high latency variance. In some example scenarios, this may entail sacrificing or risking not being able to capture liquidity at another venue.
[0186] For example, in the above situation, based on the ratio of liquidity / latency variance of venue A versus venue B or other metric(s), the processor(s) may be configured to determine timing parameters that provide a higher probability that liquidity at venue A will be captured. For example, assuming a competing trading system requires 50 μs to observe an order at venue B and modify the order at venue A, the processor(s) may be configured to determine timing parameters and / or timing sequences such that an order request is sent to venue A at time=0 and an order request is sent to venue B after time=51 μs. In this manner, if the order at venue A executes at 180 μs, which is the slowest end of its range, and the order at venue B executes at 130 μs (79 μs latency, 51 μs delay), which is the earliest end of its range, taking into account the competing trading system's 50 μs time threshold, the order at venue A could be fulfilled without any arbitrage risk from the order at venue B. As illustrated by this example, the processor(s) may be configured to target venue(s) with high latency(s) if they have large available liquidity, even though they may risk not being able to capture orders at other venue(s) with smaller available liquidity.
[0187] In this example, the processor(s) was configured to take the worst case scenario, but based on risk tolerance, the processor(s) may be configured to select more aggressive timing parameters and sequences.
[0188] In some examples, the processor(s) may be configured to select venues to target based on liquidity and latency variance. In a first scenario, if a large proposed trade can be obtained by completely avoiding venue(s) with large latency variance(s), the processor(s) may be configured to select venue(s) that do not have large latency variance and route the trade request thereto.
[0189] In some examples, the processor(s) may be configured to divide large proposed transactions between venues with large latency variances and other venue(s) with smaller latency variances.
[0190] If a venue has a large latency variance caused by deliberate or artificially created delays, in some examples, the processor(s) may be configured to target liquidity at that venue by sending multiple orders to the same venue in a timing sequence. For example, if the venue processor applies a randomly assigned delay that is large relative to typical order execution latency, in some examples, the SOR processor(s) may be configured to detect that a previously sent order is delayed and may send one or more subsequent orders in the hope that one of the orders will be processed soon. In some examples, the processor(s) may be configured to detect that a previously sent order is delayed by observing an order confirmation or response message indicating that a subsequent order to the same venue is being processed.
[0191] In some examples, multiple orders may be routed to a venue with a randomly assigned delay in conjunction with orders to other venues. In some examples, by sending multiple orders to a venue with a random delay, there may be a higher probability that at least one of the orders will be processed with little or no delay. In some examples, the processor(s) may be configured to coordinate burst orders to a venue with a random delay with orders to other venues. By using timing parameters and / or timing sequences based on a small range of intentionally induced delays for the random delay venue and execution latencies for the other venues, the processor(s) may be able to capture liquidity at both the random delay venue and the other venues. In some examples, this may reduce missed liquidity at the other venue(s) and / or increase the chances of capturing at least a portion of the liquidity at the random delay venue. In some examples, one or more of these timing and burst orders applied by the processor(s) may result in a higher capture ratio.
[0192] In some examples, burst orders to random delay venues may each be for a smaller volume to reduce the chance of exceeding the original proposed order.
[0193] In some examples, the processor(s) may be configured to determine the number and size of the orders in a burst based on the liquidity and delay range applied by the venue. In some examples, the number and size of the orders in a burst may be a function of the number of known arbitrage systems for the random delay venue, the port capacity of these systems, and / or the throttle rates of these systems.
[0194] In some examples, the processor(s) may be configured to determine the number and size of burst orders based on historical burst order performance for a particular venue.
[0195] In some examples, when most of the liquidity is available at other venues, the processor(s) may be configured to send bursts with fewer orders to the random delay venue in a timing sequence based on the short end of the possible delay range. In some cases, this can prioritize obtaining liquidity at other venues while potentially sacrificing or giving up some of the liquidity at the random delay venue.
[0196] In some embodiments, the processor(s) may be configured to send multiple orders to a single venue (random delayed or otherwise) to potentially capture reserves or iceberg liquidity that do not appear on the book.
[0197] For example, based on data related to previous orders or data received from a data source(s), the processor(s) may determine that reserve volume may be available in addition to what is visible at a particular venue.
[0198] In some examples, the processor(s) may be configured to split a larger order into smaller order requests, including two or more order requests to the same venue. In some examples, these two order requests may be routed with timing parameters and / or sequences to capture reserves or iceberg liquidity at the venue.
[0199] For example, if the processor(s) determines, based on past trading data, that a particular venue may re-post available reserves or iceberg liquidity 80 μs after the first order is filled, then the processor(s) may be configured to send two orders to the venue 80 μs apart. In one example, the processor(s) may be configured to send two orders in a sequence less than 80 μs apart if the processor(s) determines that the second order is likely to be filled regardless of whether the venue re-posts the underlying liquidity.
[0200] In another example scenario, the processor(s) may determine that venue A has 500 shares available at price X and has a latency of 30 μs, and venue B has 200 shares available at price X and has a latency of 100 μs and is likely to have underlying reserves / iceberg liquidity.
[0201] In one such case, the processor(s) may be configured to send a first order to venue B for 200 shares at time=0, a second order to venue B for 200 shares at time=50 μs (so as not to overload venue B or as reserve liquidity becomes available, or otherwise), and a third order to venue A for 500 shares at time=70 μs. In this example, the timing sequence determined by the processor(s) may enable the system to capture multiple segments of liquidity at venue B while also capturing liquidity at venue A.
[0202] In one example, the processor(s) may be configured to determine timing parameters / sequences to split the order(s) and capture multiple portions of the reserve / iceberg volume at the first venue and liquidity at the other venue(s) before the counterparty system can change volume or pricing at either the first venue (having reserve / iceberg volume) or the other venue(s).
[0203] Arbitrage and other problems caused by differences in execution times between servers can also be minimized or eliminated by reducing the absolute latency in transmitting and executing processing requests. Thus, the determination of timing parameters as described above can be implemented in combination with several procedures that also serve to minimize the absolute amount of time associated with the execution and / or reporting of execution requests by the resource(s) 106, 1106.
[0204] The determined latency information used in determining timing parameters to be associated with various portions of a multi-part execution request provided by the router(s) 104, 1104 to the multiple execution processors 106, 1106 may include timing information (e.g., transmission delays, signal propagation delays, serialization delays, queuing delays, and / or other processing delays in the router processor(s) 104, 1104, the networked computing resources 106, 1106, and / or the network(s) 110, 1110, 108, 1108). Such information may be provided by or received from any source(s) and may be stored in or retrieved from one or more data stores 214. The timing data store(s) 214 may, in various embodiments, include a database or other data structure resident in memory(s) 118, 1018 associated with or accessible by the router processor(s) 104, 1104. For example, if the execution of a portion of an execution request associated with a first networked computing resource 106, 1106 has a longer determined latency than that associated with a second networked computing resource 106, 1106 (e.g., as in the case of Exchange 1 versus Exchanges 2 and 3 in FIG. 3 ), the timing for several portions associated with the requests of a trade request to be routed to these two networked computing resources 106, 1106 may be determined such that the execution request or a portion thereof associated with the first networked computing resource 106 is sent earlier than the request associated with the second networked computing resource 106, so that the requests are executed at the two networked computing resources 106 substantially simultaneously or within an effective minimum time A or B associated with a possible condition manipulation by the trading server 304.
[0205] In some embodiments, one or more algorithms, which may employ, for example, latency probability models or other predictive models, may be used in determining timing parameters to be associated with portions of execution requests to be routed to various execution processors 106, 1106 based on information related to such communication and / or processing delays or latencies. For example, a rolling average of historical latency data accumulated or associated over any desired devices, time periods, or other timing considerations may be used to predict expected latency for the execution of data processing requests.
[0206] One example of an algorithm suitable for use in determining timing parameters to be associated by the router(s) 104, 1104 with the portion(s) of a request for execution provided by the source(s) 102, 1102 is based on the average latency between transmission of a request signal from the router(s) 104, 1104 and an appropriate timing reference if it is desired to effect simultaneous or synchronized arrival of such portions of requests at the network resources 106, 1106. Such timing reference(s) may include, for example, initiation of processing by the corresponding targeted resource(s) 106, 1106 and / or receipt by the routing processor(s) 104, 1104 of an acknowledgement signal generated by the resource(s) 106, 1106 upon receipt of the request and / or completion of execution of the request. For example, in some embodiments, it may be advantageous to measure the latency between a transmission to a given resource 106, 1106 and receipt by the router(s) 104, 1104 of an acknowledgment or acknowledgement signal or other suitable response signal 1260 from such resource 106, 1106, and to use such measured latency(s) in determining 210 the timing parameter(s).
[0207] Process step 210 may be performed, for example, by an application executed by or associated with routing processor(s) 104, 1104, such as capital management entity or module 1126 in the case of financial system 1000. Determining the timing parameters to be associated with each part or segment of a multi-part execution request may include, for example, use of an adaptive exchange round trip latency (RTL) learning and compensation logic module 1126c, such as that shown in FIG. 1B. With reference to FIG. 3, such an adaptive exchange RTL learning and compensation logic module 1126c may determine the timing for each signal processing request (e.g., trade request) as follows: 1) For each portion or segment n of a multipart processing request X of m parts, a time T1 provided by a clock associated with, for example, the processor(s) 104, 1104 x,n is time-stamped by the processor(s) 104, 1104 at a desired defined point in the parsing or generation process of the trade order(s) or other processing request(s) X and is associated with the processing request signal set record(s) corresponding to each part or segment n of the m-part request X. 2) T2 for each portion n of multipart request X x,n is time stamped by the processor(s) 104, 1104 when the corresponding Nth portion of the request signal set is received at the targeted exchange 106, 1106 and the corresponding exchange generated confirmation message is received by the requesting routing processor 104, 1104. 3) During the course of a trade date (or other data processing period), process steps 2 and 3 are repeated to determine the corresponding T1 x,n and T2 x,n may be routed to a given execution processor 106, 1106. 4) For each portion segment n of a subsequent pending multipart execution request Y, the timing parameters determined are y,n =Σ(T2 x,n -T1 x,n ) / Z, where Z is the number of previously executed order segments routed to a given execution processor 106, 1106 used in the calculation.
[0208] Timing data store(s) 214 may store historical timing parameters (e.g., a plurality of determined timing parameters RTL) associated with one or more execution resources 106 / exchange servers 1106. y,n ), such data may be used to create a rolling histogram that can be used to predict the current or cumulative latency for each resource 106 / exchange server 1106. Because such predictions are based on continuously changing ("rolling") records, this process may be referred to as "online learning." A component responsible for this (e.g., an exchange latency histogram memory or processing component, not shown) may reside within the adaptive exchange RTL learning and compensation logic module 1126c.
[0209] The adaptive exchange RTL learning and compensation logic module 1126c may use the predicted latency to determine appropriate timing parameters to be used in transmitting trade (or other data processing) requests to various exchange servers 1106 to reduce, control, minimize, or eliminate timing differences in execution of portions of split trade requests routed to different exchange servers 1106, and thus compensate for differences in execution latency associated with such exchange servers 1106 in a manner that reduces or eliminates arbitrage in latency by opportunistic traders, for example.
[0210] The adaptive RTL module(s) 1126c may use various algorithms in determining appropriate timing parameters to use in synchronizing the execution of a multi-part processing request. For example, such a module may use the latency values determined for various exchanges to determine the extent to which the router(s) 104, 1104 should compensate for different exchange latencies, for example by sending their corresponding portions of the request for processing at different times to the various processors 106, 1106. This may minimize delays in the completion of the execution of each portion, for example by minimizing the time difference in the receipt of each respective portion by its corresponding execution resource 106, 1106. (In FIG. 3, for example, this would be shown as minimizing the difference in the time elapsed at time X, time Y, and time Z). Such an algorithm may also take into account historical differences in the time it takes to execute a trade or other processing order on the various resources 106, 1106, in addition to communication delays. The adaptive RTL module(s) 1126c may be configured to determine timing ranges to maintain, for example, a network monitoring model, a network monitoring log, and the like.
[0211] The adaptive RTL module(s) 1126c may also be configured to automatically adjust timing parameters / sequences depending on information such as execution rates, trade-win ratios, predicted network characteristics, and the like.
[0212] The adaptive exchange RTL learning and compensation logic module(s) 1126c may further collect information about the prevailing market conditions at each exchange server 1106 (e.g., using sources of data such as exchange market data source(s) 1126v), wave orders / executions, actual latencies, and target latencies (e.g., as predicted above) when a trade request is sent. A component responsible for this may reside within the adaptive exchange RTL learning and compensation logic module 1126c.
[0213] One or more timing parameters associated with an execution request to be routed to any one or more of the execution processor(s) 106, 1106 may also be provided to the corresponding routing processor(s) 104, 1104 (e.g., in the timing data store 214) or determined by such processor(s) 104, 1104 using related data provided by any one or more market data feed(s) or processor(s) 1126 (e.g., including any one or more of the processors or (sub)systems 1126a-1126g and / or 1126v) and / or by the processor(s) 106, 1106 themselves.
[0214] In some examples, the processor(s) may be configured to evolve by randomly introducing small variations in various timing parameters and / or venue selection factors. By measuring the resulting winning ratios over time and readjusting the timing parameters and / or venue selection factors, the processor(s) may in some examples evolve toward a more optimal timing and / or factor weighting. In some examples, this feedback loop may enable the processor(s) to automatically adjust for any changes in the network, venues, competitor behavior, etc.
[0215] In some examples, by considering the win ratio or execution rate, the processor(s) can essentially take into account any number of underlying factors, such as network conditions, networked resource conditions, competitor behavior, etc., without a specific understanding of which underlying factors may have affected the win ratio. In some cases, this may result in a simpler and / or more efficient system that can adjust for changes or variations without a potentially complex calculation of every possible underlying factor.
[0216] At 212, the various portions of the optionally aggregated and partitioned signal processing execution request(s) are sent to the respective networked computing resources 106 according to the timing parameters or sequence(s) determined or otherwise obtained at 210. The request(s) or various portions thereof may then be executed by the respective execution resources 106, 1106, along with subsequent signal communication and processing as needed or desired. As will be understood by those skilled in the art, once the parameters of the desired execution request have been determined by the router(s) 104, 1104, and the skilled artisan is familiar with this disclosure, signals representing these parameters may be assembled using known or specialized data processing techniques; formatted according to the Financial Information Exchange (FIX) protocol and / or any other desired protocol(s); and transmitted, written to, or otherwise communicated to the corresponding execution processor(s) 106, 1106 using known or specialized signal communication techniques, and executed according to the requested trade or other data processing.
[0217] For example, continuing with the above example, the timing delays or one or all of parameters X′, Y′, Z′ may be equal to zero or any other suitable time period and may be determined in accordance with the disclosure above and associated with order segments generated by processor(s) 1104 for the purchase of 77,000 bond lots of CUSIP No. AA Notes at price A with 25,000 lots (18,000+7,000) in reserve at prices D and E, respectively, as follows: <delay X’><exchange A1> <sell><CUSIP No.AA><15,000><price A> <res.6,000><price D><res.2,000><price E> <delay Y’><exchange B2> <sell><CUSIP No.AA><27,000><price A> <res.6,000><price D><res.2,500><price E> <delay Z’><exchange C3> <sell><CUSIP No.AA><35,000><price A> <res.6,000><price D><res.2,500><price E>
[0218] The routing processor(s) 104, 1104 can then process the trade segments by using timing parameters, e.g., delay X', Y', Z', to result in the corresponding trade segments to be transmitted or otherwise provided to the exchanges 106, 1106, A1, B2, C3 for execution according to a desired timing sequence for simultaneous or other desired sequential execution.
[0219] After execution of all or a substantial portion of the routed transaction or processing segment, the routing processor(s) 104, 1104 may receive data confirming or otherwise indicating such execution from the corresponding execution processor(s) 106, 1106 and may assign the execution results to the requesting source(s) 102, 1102 by accessing data records stored in the associated memory(s).
[0220] Reference is now made to Figure 4, which illustrates an example method 300 for determining timing parameters to be used in managing data processing by multiple networked computing resources 106. In the illustrated embodiment, the method 300 is iterative, with each loop of the method 300 being denoted as N. The method 300 is suitable for implementation using any of the various embodiments of the systems 100, 1000 and components thereof, including, for example, the router processor(s) 104, 1104 and the data source(s) 1126, among others.
[0221] At 302, each of the plurality of networked computing resources 106, 1106 is monitored for latency associated with receiving and / or executing signal processing execution requests, for example, by the router processor(s) 104, 1104, the execution processor(s) 106, 1106, the external processor(s) 1126, and / or various components or modules operated by or associated with it. This may be performed, for example, by a monitoring module in the router processor(s) 104, 1104 (e.g., exchange RTL measurement module 1126b for the financial system 1000, etc.). Such monitoring may include, for example, time-stamping outgoing requests for data processing and comparing the time of receipt of confirmation(s) or results from processing with the corresponding time-stamped outgoing requests. The time difference between an outgoing request and an incoming acknowledgement and / or data processing result may be defined as data or signal processing latency and stored in memory accessible by the router processor(s) 104, 1104. Such latency may be monitored on a continuous, periodic, and / or other dynamic basis depending on the timing difference between an outgoing request and an incoming acknowledgement, acknowledgement, and / or result.
[0222] At 306, at least one timing parameter is determined that relates to latency(s) observed in the execution of signal processing requests provided to the resources 106, 1106 monitored by the routing processor(s) 104, 1104. As described herein, such timing parameter(s) may include, for example, latency due to communication delays, such as transmission delays or other signal propagation delays, and / or processing delays, among others. Typically, a corresponding timing parameter(s) is determined for each of a plurality of networked computing resources 106, 1106 to which a trade order or other data processing request or portion thereof is expected to be sent by the routing processor(s) 104, 1104.
[0223] In various embodiments, such as the various forms of financial system 1000, depending on the type of system(s) used and the desired processing results, such timing parameters may be determined with respect to one-way and / or round-trip communication between the routing processor(s) 1104 operated by or on behalf of the capital management entity and the exchange server 1106, i.e., from the generation of a multi-part trade request by the capital management entity's routing processor 1104 to the receipt of a response, such as confirmation of receipt of a portion of a larger trading order from the execution resource to which the processing request was directed and / or confirmation of execution of all or a portion of the requested trade. 1B , for example, as described above, the RTL measurements may include latencies due to any or all of the following: transmission of signals within the capital management entity server 1104, processing of signals within the capital management entity 1104, transmission of signals between the capital management entity 1104 and the network 1110, transmission of signals within the network 1110, transmission of signals between the network 1110 and the targeted exchange server 1106, and processing of signals within the exchange server 1106 for both communications sent from the routing processor(s) 104, 1104 and responses sent from the exchange server 106, 1106 (e.g., acknowledgment of the communication, rejection of the trade request, confirmation of the trade request, etc.). In such an embodiment, the timing parameter(s) may simply be the total time for the round-trip communication or a statistical or other mathematical function thereof.
[0224] For example, an exchange RTL measurement module 1126b, such as the one associated with the SOR 1104 shown in FIG. 1B, may determine the timing parameters as follows: 1) A timestamp value T1 is associated with a new communication M1 (eg, a trade request) sent by the processor(s) 1104 to the exchange server 1106. 2) The timestamp value T2 is associated by the processor(s) 1104 with any response to request M1 received from the exchange processor 1106 to which request M1 was sent. This response may be any response, such as an acknowledgement, a rejection, a full or partial fulfillment, etc., and may depend on the nature of the request represented by M1. 3) The RTL associated with request M1 is calculated as the difference between T2 and T1. In one embodiment, as mentioned above, the RTL may be calculated as the average of the time (T2-T1) for the past number Z (e.g., 30) of processing requests routed to each of the multiple targeted exchange processor(s) 1106.
[0225] At 308, the timing parameter(s) associated with each networked computing resource 106 may be stored in timing data store(s) 214. As described herein, the timing data store 214 may, in some examples, be a database or other data structure resident in memory associated with or accessible by the router processor(s) 104. The timing parameter(s) stored in the timing data store(s) 214 may be used in processes such as those described above in connection with process block 210 of FIG.
[0226] The timing parameter(s) determined by the processor(s) 104, 1104 may represent, for example, a rolling histogram(s) representative of latencies associated with individual executing processors 106, 1106 and / or other components of the system(s) 100, 1000.
[0227] FIG. 5 shows an example histogram illustrating stored data representing processing latency values associated with communications and / or other processing involving the execution processors 106, 1106 in the system 100, 1000. In the illustrated example, round trip latencies (ms) are stored for the last 30 trade requests or other communications with a given execution server 106. Although this example shows 30 latencies stored, the number of stored timing parameter(s) used in determining the RTL or other timing parameters may be more or less and may vary according to conditions such as time of day, season, etc. Results of calculations based on the stored latencies and other related data may also be stored in the timing data store(s) 214. For example, in the example of FIG. 5, in addition to the raw latencies, a rolling average or rolling mode of the last 30 (or other suitable number) latencies associated with communications and / or other processing with or by each execution server 106 may also be calculated and stored in the timing data store(s) 214.
[0228] As will be readily understood by those skilled in the art, additional factors including, for example, a desired fixed offset or delay, or scaling factors related to known trading or other data processing patterns, economic conditions, etc., such as time of day, day of the week, season, etc., may be used in determining the timing parameters at 210.
[0229] The timing parameters determined in 210 may be used by the routing processor(s) 104, 1104 to synchronize execution of processing requests generated by the source(s) 102, 1102 and directed to the processor(s) 106, 1106, for example, by associating such requests, or portions thereof, to be forwarded for execution by each of the multiple processor(s) 106, 1106 with data items usable by the processor(s) 104, 1104 to effect communication of the requests to the corresponding processor(s) 106, 1106 at a desired absolute or relative time to achieve a desired synchronization of arrival of the requests at the corresponding executing processor(s) 106, 1106. For example, by using data items configured to cause communication of one or more portions of a request at a given time(s) according to a clock associated with the processor(s) 104, 1104, the processor(s) 104, 1104 can cause the request(s) or portion(s) of a request to be communicated at a desired date and time, or in any desired relative order or sequence that is not related to the actual date and time, but rather to each other or some third index.
[0230] At 310, N is incremented by 1 or other suitable value, or control is returned to 302 so that processes 302-308 continue. Optionally, processes 302-310 continue until a maximum desired number of iterations have been completed, or until all requests for trading or other processing by the order have been processed (e.g., routed to an execution processor 106, 1106), or other suitable criteria have been met.
[0231] To assist operators and users of system(s) 100, 1000 or components thereof in understanding or evaluating the effectiveness of the disclosed methods and systems for effecting data processing by multiple networked computing resources, in certain aspects the present disclosure also provides various metrics (e.g., trading benchmarks, in the case of financial system 1000) that may be determined by and through the use of data generated from any or all of the various components of system 100, 1000.
[0232] Reference is now made to FIG. 6, which illustrates a comparison of the results of transmitting a multi-part trade execution request to multiple networked computing resources or execution processors 106, 1106 in accordance with an example of the disclosed method and system, versus the results of a multi-part trade request transmitted conventionally.
[0233] 6A illustrates the results of execution of a multi-part trade request using the disclosed method and system to obtain synchronized (in the illustrated case substantially simultaneous) execution of various parts or segments 624 of the multi-part trade request (sell order) by multiple exchange servers 106, 1106. In the illustrated example, a 94% execution rate of the original aggregated order was achieved at an original offer price 630 of $4.21 (shown as "Level 1"). In a second round of trading (executed in a single trade as shown at 626), the remaining volume was sold at a less desirable but still acceptable price 632 of $4.20 (shown as "Level 2"). The costs associated with orders that were executed below the requested order price (i.e., those orders at Level 2) were $53,000 for the trader system 1102 (e.g., client system) and $10,049 for the capital management entity 1106.
[0234] In Figure 6B, using prior art trading methods and systems, an unsynchronized multi-part trade request (multi-exchange sell orders) consisting of multiple unsynchronized order segments 624' for the same overall trade request resulted in an initial execution rate of 47% at a preferred order price 630 of $4.21 (shown as "Level 1"). An additional 43% of the request was then executed at a less desirable price 632 of $4.20 (shown as "Level 2"), with the remainder executed at a further reduced price 634 of $4.19 (shown as "Level 3").
[0235] Using the method and system of the present disclosure, a volume weighted average ask price (VWAP) 636 of $4.2094 / share was achieved, as shown at 628. Using the prior art method and system, a VWAP 638 of $4.2038 / share was achieved.
[0236] As will be readily appreciated by those skilled in the art, the system 100, 1000 may include devices or components suitable for providing a variety of additional metrics and functionality. Reference is now made to FIG. 7, which illustrates two examples of the provision by the routing processor 104, 1104 or other processor of benchmark comparisons against market average prices, for example, as provided by a market news service or other market data source 1126v. At 646, the performance of the system 100, 1000 in the synchronized processing of multi-part trade requests according to the present invention is compared to a market performance indicator "average price benchmark." Such average price benchmark or other benchmark or metric factors may be obtained, for example, from any or all of the components 1126, 1106. At 644, the performance of the system 100, 1000 in the non-synchronized processing of multi-part trade requests according to the prior art method is compared to the same market performance indicator "average price benchmark." The comparison of the comparisons 646, 644 indicates that the processing of transactions according to the present invention results in better outcomes for sellers of financial rights. As will be appreciated by those skilled in the art, a variety of benchmarks may be used in evaluating the performance of the systems and methods of the present invention, which may depend at least in part on the nature of the system 100, 1000 being used and the type of trades or other execution requests being processed by such system.
[0237] 8 is a flow chart illustrating an embodiment of an example method 800 for coordinating data processing by multiple networked computing resources. The method is suitable for implementation by, for example, the router processor(s) 104, such as the SOR 1104, or by any one or more processors in the system 1000.
[0238] Aspects of the exemplary method 800 in Figure 8 are similar or the same as those seen in Figures 2 and 4 and described in various exemplary embodiments described herein. Thus, any example or implementation details described with respect to these figures or described herein may be applied to the method of Figure 8. Similarly, examples and implementation details described with respect to Figure 8 may be similarly applied to the exemplary methods of Figures 2 and 4. All of these variations and combinations of these aspects are contemplated by the present description.
[0239] At 810, one or more processors 104 in the system 1000 monitor data associated with the networked computing resources 106. As described herein, in some embodiments, the data associated with the networked computing resources may include, among other things, data associated with data processing segments previously routed to the networked computing resources.
[0240] In some embodiments, data relating to previously routed data processing segments may include timing information relating to when the data processing segments are routed (e.g., using a timestamp), when the data processing segments are processed at the networked computing resources (e.g., a timestamp or processing time field in a response message), and when the response message is received at the system (e.g., using a timestamp). In some embodiments, monitoring such data may include determining execution latencies (and / or different latency components) for the data processing segments and associating them with corresponding networked computing resources.
[0241] In some embodiments, monitoring the data may include monitoring and comparing latencies for different types of data processing segments. For example, different types of data processing segments routed to the same venue may have similar transmission latencies, but different latencies associated with the actual execution of the data processing segments. In another example, data processing segments having different lengths, or even no instructions, may have different latencies when routed to the same networked computing resource.
[0242] In another example, improperly formatted data processing segments may be quickly rejected by the networked computing resource such that the processor may be configured to attribute all or most of the latency associated with such data processing segments to transmission latency.
[0243] In one embodiment, a processor in the system is configured to determine different latency components associated with a particular networked computing resource based on differences in total latency for different order types / lengths / etc.
[0244] In some examples, the latency components may include, but are not necessarily limited to, send transmission latency, execution latency, and reply latency (eg, of a response message).
[0245] The monitored data related to a previously routed data processing segment may, in some examples, include data in one or more response messages indicating whether and / or to what extent the data processing segment was successfully processed. For example, data in one or more response messages to a trade request, a data processing segment representing a trade request, may indicate whether the trade was fulfilled, how much of the request was fulfilled, and / or the price at which the request was fulfilled.
[0246] The monitored data relating to previously routed data processing segments may include data indicative of liquidity of financial rights posted or known to be available on the networked computer resources at the time the first data processing segment in the wave is routed or at a time prior to the first routing.
[0247] In an embodiment, monitoring the data associated with the previously routed data processing segments may include determining one or more capture ratios for each data processing segment generally and / or for one or more data processes. The processor(s) may determine the capture ratio by comparing available liquidity targeting the networked computing resources to the actual liquidity captured by the trade request. The capture ratios of the previously routed data processing segments may be monitored on a segment-by-segment basis across all data processing segments split off from the original one or more data processes.
[0248] In some embodiments, the monitored data may include data parameters that identify a risk of information leakage for each of the networked computing resources. For example, the parameters may be set to indicate the presence of a co-located or active low latency third party server, a rebate scheme that encourages one or more trade request types, data related to data processing segments previously routed to the networked computing resource indicating that one or more related data processing segments were unsuccessful, etc. In some examples, monitoring the data includes generating a risk score of information leakage based on the monitored data.
[0249] In some embodiments, the system has multiple network routes to a networked computing resource, and in these embodiments, the monitored data associated with a particular networked computing resource may be monitored / associated on a per route basis.
[0250] In addition to routes, in some examples, the monitored data may also include latency data and / or status information for routing a segment(s) along one or more routes and / or a device(s) on one or more routes.
[0251] In some embodiments, monitoring of data may include collecting, measuring, or requesting data from one or more components or devices in the system as described herein or otherwise. Monitoring of some data may include various calculations.
[0252] Monitoring the data may include storing the data in one or more memories or data storage devices in the system. In some embodiments, multiple data points associated with the networked computing resources may be collected to provide a range and / or distribution / probability of the data.
[0253] At 820, the processor(s) receive signals from one or more data sources 102 representing instructions for execution of one or more data operations executable by one or more of the networked computing resources 106. As discussed above, for example, with reference to FIGURE 2, the one or more data operations may, in some embodiments, represent a request for execution of a trade in financial rights and / or other transaction.
[0254] In some examples, instructions for performing the data processing(s) may include specific parameters for performing the data processing, including process priority, tolerance values for information leakage, maximum acceptable latency between any data processing segments, desired routing paths, desired venues, specific routing instructions, etc.
[0255] At 830, based on the monitored data, the processor(s) create data processing segments for the one or more data processes. In some examples, creating the data processing segments may include combining, dividing, or otherwise determining and creating a number and size of data processing segments for performance of the received one or more data processes.
[0256] In one embodiment, creating or partitioning at least one data processing segment of the data processing of the processor(s) includes selecting a networked computing resource to route the one or more data processing segments to, the selection being based on monitored data associated with the networked computing resource.
[0257] The data processing creation / partitioning also includes determining a size of each of the data processing segments. In some embodiments, this may include a size of a processing task or a size of a trade request. The processor is configured to determine a size of each of the data processing segments based on the monitored data.
[0258] The selection and sizing decisions, in some embodiments, may be based on monitored data including one or more of the liquidity available at the networked computing resource for the financial right(s) associated with the data processing, the risk of information leakage, and the latency time(s) associated with the networked computing resource.
[0259] The creation of a data processing segment can be based on available liquidity for targeting across multiple networked computing resources, but in some instances the processor(s) may determine that a single data processing segment should be sent to a single networked computing resource. Alternatively, this may be viewed as sending a size zero data processing segment to other networked computing resources, or not sending a data processing segment at all.
[0260] At 840, the processor(s) determine timing parameters for the data processing segments. The timing parameters may specify a timing difference for initiating routing of the data processing segments based on the monitored data. In one example, the timing parameters may define a timing offset or delay between a time at which the first data processing segment should be routed and a routing time of the second data processing segment. In one example, the timing parameters may define a range between a time at which the first data processing segment should be routed and a routing time of the second data processing segment.
[0261] In some embodiments, the timing parameters may define probabilities or risk factors of information leakage associated with different time subranges within the time range.
[0262] The timing parameters are determined to provide for synchronized execution of the data processing segments. In some examples, the timing parameters may be determined to aim to execute the data processing segments as close as possible to one another. In some examples, the timing parameters may be determined to aim to execute the data processing segments in a desired sequence and / or with a desired relative timing. In some examples, the timing parameters may be determined to aim to capture as much liquidity as possible and / or at the most desirable price. In some examples, the timing parameters may be determined based on a distribution of randomly introduced delays at one or more of the networked computing resources.
[0263] At 850, the processor(s) begin routing the data processing segments to their respective networking computing resources based on the timing parameters.
[0264] As described herein or otherwise, the determination of the data processing partitioning, number, size, and destination networked computing resources of the data processing segments is based on monitored data associated with the networked computing resources.
[0265] In some circumstances, the processor may be configured to consider different trade-offs between available liquidity, risk of information leakage, and / or latency distribution for one or more networked computing resources.
[0266] For example, by default or based on a risk tolerance value associated with the data source from which the data processing is being received, the processor(s) may create data processing segments and timing parameters to target as much liquidity as possible, albeit with a higher risk of data leakage and not obtaining a portion of the liquidity or a better price. Conversely, by default or based on a different, lower risk tolerance value associated with the data source, the processor(s) may create data processing segments and timing parameters to target a smaller amount of liquidity with a lower risk of data leakage.
[0267] As described herein or otherwise, in some embodiments, a processor may create data processing segments and timing parameters based on available liquidity.
[0268] For example, by continuously and / or periodically monitoring data associated with networked computing resources as illustrated in Figure 4, the system may, in some cases, establish more accurate and / or up-to-date data regarding latency, risk factors, liquidity, etc. In some examples, over time, the monitored data may establish the frequency, distribution, etc. of one or more types of monitored data.
[0269] In some examples, by monitoring timing parameters and result data for previously performed data processing segments and waves, the processor may establish a local minimum or maximum that provides a better outcome. In some embodiments, when determining and / or applying timing parameters for routing, the processor(s) may introduce random timing variations into the timing sequence. In some examples, these variations may be small relative to the timing parameters and / or may be selected such that the changed times still fall within a range that satisfies the timing parameters. In some examples, by introducing timing variations and monitoring data associated with the affected data processing segments, the resulting adjustments to the timing parameters may converge more quickly to a more optimal value or may find a different local minimum / maximum.
[0270] In some cases, the timing parameter distribution may be multimodal, for example, when a particular networked computing resource randomly introduces delays or speed bumps. In some embodiments, the processor may divide the data processing into at least two data processing segments to be routed to a particular networked computing resource and establish timing parameters based on the multimodal distribution. For example, the processor may generate bursts of data processing segments with smaller trade requests with the goal of allowing at least a subset of the data processing segments to execute without incurring intentional delays or speed bumps at the networked computing resource.
[0271] 1B, the source(s) of monitored data 1126 usable by the processor(s) 104 in generating financial trade or other data processing execution requests includes a number of modules 1126a-g useful in generating multipart execution requests. In the illustrated example, the modules 1126a-g include a market data processing module 1126a, an exchange round trip latency measurement module 1126b, an adaptive exchange round trip latency (RTL) learning and compensation logic module 1126c, a smart sweeping stock allocation logic module 1126d, a smart posting logic module 1126e, a regional and national exchange access logic module 1126f, and an aggressiveness management module 1126g.
[0272] Market data processing module 1126a receives and processes market data, which may be the same or different from the data provided through exchange market data module 1126v of exchange server 1106. Sources of such data may be internal or external to system 1104 as needed or desired, and may include any suitable privately or publicly available source of data useful in creating execution requests, particularly requests useful in splitting or otherwise creating trade orders, where the information provided may include, for example, numbers or quantities and / or prices available on any particular exchange; historical trading volume or prices; current and historical depth of the market(s) or liquidity; reserve sizes; absolute, relative, and / or average price spreads; and stock or right specific heuristics; and / or trends in any or all of the above.
[0273] The exchange RTL measurement module 1126b determines timing parameters for use in synchronizing the execution of multi-part trades or other data processing requests by multiple exchange servers 1106, for example, using statically defined latency data representing the time(s) elapsed between the sending of a request or other data to an individual execution processor(s) 106, 1106 and the receipt of a confirmation or execution result from the individual execution processor(s) 106, 1106, as described herein.
[0274] The adaptive exchange RTL measurement module 1126c determines timing parameters for use in synchronizing the execution of multi-part trades or other data processing requests by multiple exchange servers 1106, e.g., using dynamically defined ("rolling") latency data representing the time elapsed between the transmission of multiple processing requests or other data to an individual execution processor(s) 106, 1106 and the receipt of a confirmation or execution result from the individual execution processor(s) 106, 1106, as described herein. Histograms and other data models and / or structures representing such rolling data may be used by the module(s) 1126c in determining timing parameters according to such a process.
[0275] The smart sweeping stock allocation logic module 1126d includes statistical models for strategically oversizing trade requests and / or associating reserve quantities with publicly posted orders based on historically observed market data. This module 1126d determines the appropriate oversizing (i.e., over-ordering on trade requests) to be incorporated into open orders, taking into account the predicted amount(s) of illegitimate assets at the exchange server 1106, for example, based on statistical data about illegitimate assets available at the exchange server 1106 over a given period of time or under other specified conditions (e.g., the past 30 trade requests). Based on such predicted hidden market reserves, appropriately sized illegitimate assets can be determined and associated with trade orders, resulting in strategic oversizing of publicly visible orders and helping to ensure that actual desired trading volumes are realized.
[0276] The smart posting logic module 1126e includes statistical models for determining the probability of execution (i.e., the satisfaction percentage of trade requests) expected to be achieved on trade requests routed to individual exchange servers 1106. Such statistical models may include, for example, historical execution data achieved on such individual exchanges over a given period (e.g., the last 30 trade requests, the previous month, the previous 12 months, etc.). The smart posting logic module 1126e may take into account factors including, for example, the top-of-book depth at each exchange server 1106, the volatility level across the exchange servers 1106, and the average latency to execution of trade requests, among others.
[0277] The smart posting logic module 1126e may also be configured to determine a trade capture ratio for the historical order, for example, by dividing the desired liquidity at the time the order was submitted and / or transmitted with the actual captured liquidity achieved through execution of the order.
[0278] The regional and national exchange access logic module 1126f provides information on how trade requests are routed to the exchange server 1106 depending on whether the exchange server 1106 is regional or national. Internally and / or externally stored data related to the appropriate protocol(s) to be used, regulations observed, etc. may be used in providing such data. Such data may be used, for example, in ensuring that trades or other processing requests forwarded by the routing processor(s) 104, 1104 to the external resource 106, 1106 are properly formatted from the perspective of the resource(s) 106, 1106 to which the request(s) are provided, and in ensuring that such request(s) comply with all applicable legal standards.
[0279] The challenge management logic module 1126g includes a probabilistic model for determining execution percentage probabilities for individual exchange servers 1106 and appropriately modifying execution requests routed to such servers. Such a module 1126g may take into account factors such as, for example, the execution rate at each exchange server 1106, the book depth at each exchange server 1106, and the volatility level across the exchange servers 1106, among others.
[0280] In some embodiments, specially constructed order messages may be used for routing orders. For example, the router processor 1104 may be for message handling of FIX messages (or other protocols). The venue side of the system performs the scheduling portion of the processing and the following message handling functions:
[0281] Adjusting the sequence number (since the sequence number can only be determined when the order is scheduled). To do this, the FIX engine will generate a constant length sequence number field that is always null (such as 8 or 9 zeros "000000000"), and then the scheduler will overwrite this field with the actual sequence number.
[0282] (To reduce processing, the FIX engine will pass along messages with the correct checksum for the "000000000" sequence number field.) The scheduler then subtracts any non-zero characters in the sequence number from the checksum when it updates the sequence number, thus avoiding recalculating the checksum for the entire message and reducing processing to a very simple mechanical operation.
[0283] The sequence number will be returned to the sender's FIX engine in the router processor 1104 so that it can be incorporated into the session log / state.
[0284] Messages from the venue will bypass the scheduler and be handled directly by the FIX engine.
[0285] In some embodiments, scheduling and message updating can be reduced to highly "mechanical" operations that are amenable to high-speed coding techniques and possible hardware acceleration (eg, in an FPGA).
[0286] As mentioned above, in some embodiments, the sequence in which order waves are removed from a queue and routed to associated venues may be based on any of the factors described herein or in other ways.
[0287] Routing trade requests in a timed sequence may involve delaying one or more trade requests in a wave / sequence. In some cases, this may improve the win ratio of a trade, but routing several trade requests in a timed sequence may inherently take longer than sending them all at once.
[0288] In some examples, the processor(s) may be configured to determine a sequence for routing multiple waves of trade requests at 208 and / or 210. In some examples, this may include interspersing, overlapping, or otherwise routing at least some trade requests for different waves simultaneously.
[0289] In one example, the processor(s) may be configured to sequence multiple waves of trade requests while respecting intra-wave timing parameters, such as minimum message intervals, as well as general timing parameters that must be respected to avoid overloading or queuing at the venue.
[0290] In some examples, the general or venue-specific timing parameters may be based on observed timing factors such as latency and performance rates. In some examples, the general or venue-specific timing parameters may be based on characteristics associated with network devices, links, and / or venues. These characteristics may include, but are not limited to, device, link, and / or performance factors based on throughput or performance limitations, service level agreements.
[0291] For example, the table of FIG. 10 illustrates exemplary monitored data received or determined from monitoring network performance, historical orders, and / or historical order data.
[0292] With these example latencies and throughputs, the example scenario involves the receipt of four order waves as illustrated in FIG. 11A (and associated default timing parameters displayed).
[0293] Without corrections, and assuming a processing rate of 10 microseconds between waves, the table in FIG. 11B lists the time when the order is routed and, in parentheses, the time when the order hits the venue processor(s) and the resulting timing error.
[0294] In this example, if the timing error must be within 100 microseconds of the target, each of waves B, C, and D would not meet the target. This scenario requires 2130 microseconds of router time and 2910 microseconds including the venue.
[0295] In another example scenario, waves C and D can be reordered to be sent earlier because they relate to unrelated symbols. The table in Figure 11C shows the timing when wave D starts immediately after wave A, and then waves B and C start simultaneously.
[0296] Waves B, C, and D all miss the 100 microsecond target, but there is an improvement in the overall time to handle all four waves (1310 microseconds at the router and 2600 microseconds including the venue).
[0297] In the next scenario illustrated in FIG. 11D, waves are routed to preserve the sequence of ordered waves, anticipating congestion in the venue.
[0298] As illustrated, all waves meet the 100 microsecond target, but the handling time has increased to 3700 microseconds.
[0299] As illustrated in FIG. 11E, in the following scenario, the wave timing handler can be configured to schedule orders to be routed in an inter-wave timing sequence as well as an intra-wave timing sequence based on both venue throughput / congestion and latency.
[0300] In the above scenario, all waves meet the 100 microsecond target and handling times are 2300 microseconds for the router and venue.
[0301] In one example, the wave timing handler may be configured to determine the optimal or best available schedule for the next N waves. In one example, when N is small (e.g., less than 10), the wave timing handler may use a brute force algorithm by computing all possible sequences with resulting handling times and selecting the best one.
[0302] In some examples, when N is larger or when a brute force approach would result in excessive delays in the wave routing process, the wave timing handler may be configured to calculate as many sequences as possible in a defined / allocated amount of computation time / cycles and select the best one.
[0303] In some examples, the processor(s) may be configured to apply other job scheduling optimization algorithms to schedule different waves within the shortest handling times.
[0304] In some examples, the number of waves to be included in a scheduling algorithm may depend on the number of venues and / or trade requests in the waves.
[0305] 12 is a flow chart illustrating aspects of an example method 1200 for coordinating data processing by multiple networked computing resources. At 1210, a processor retrieves a data processing wave from a wave or session queue. The data processing wave identifies one or more data processing segments and corresponding networked computing resources to which each of the data processing segments should be routed. The data processing wave also identifies a timing sequence in which the one or more data processing segments should be routed.
[0306] At 1220, the processor obtains a minimum handling interval for each of the networked computing resources. In an embodiment, the handling interval may be based on monitored data associated with the networked computing resources.
[0307] At 1230, the processor schedules orders for routing the data processing waves based on the wave timing sequences and handling intervals for the networked computing resources. In one embodiment, scheduling the orders includes determining orders that intersperse data processing segments of different data processing waves without disrupting the handling intervals of the networked computing resources and without disrupting the timing sequences of the data processing waves.
[0308] In some examples, the timing sequence may include acceptable time ranges between the time when a first data processing segment can be routed and the routing time of a second data processing segment. In some embodiments, scheduling the order for routing data processing waves may include adjusting the routing time sequences for one or more waves within these time ranges.
[0309] In one embodiment, the processor is configured to retrieve M data processing waves from the wave queue and determine a total handling time for each possible arrangement of the M data processing waves. The processor then schedules the order based on the arrangement having the shortest total handling time.
[0310] In some embodiments, the processor may consider additional parameters. For example, when a choice is available, the processor may be configured to select an arrangement that uses the fewest number of networked computing resource (venue) sessions. In another example, the processor may consider whether different data processing segments are trading requests regarding the same financial interest, and may select an arrangement to avoid or reduce any conflicts. Other similar factors include order size, price, parties to the trade, etc.
[0311] The selection of the number M of data processing waves may depend on the amount of time required to evaluate all order sequences for the M waves relative to the total handling time required by the order wave processor to route the waves. If the evaluation time is too high, it may become a bottleneck and waste routing resources by making them idle while the evaluation is taking place. It may also increase the amount of time during which price or liquidity at the networked computing resources may change, thereby potentially reducing the capture ratio. If M is too small, routing resources are wasted as there may not be much opportunity to optimize the scheduling of the waves.
[0312] In some examples, the number of waves M can be dynamically selected as resources and / or wave handling times change.
[0313] In one embodiment, the processor computes as many sequences for the M waves as can be evaluated before the time limit expires. Once the time limit expires, the best schedule from the sequences evaluated is selected.
[0314] In one embodiment, the processor is configured to select and schedule M waves from the wave queue, however, only the first W waves from a scheduled order are routed while the remaining MW waves in the schedule and the next W waves in the wave queue are evaluated to determine the next scheduled order.
[0315] In some embodiments, the system routes one scheduled order (or a subset of orders) of a wave while simultaneously calculating the schedule for the next order. Pipelining this process can lead to improved performance in some instances.
[0316] In some examples, the selection of waves from the wave queues and / or session queues may be based on one or more priority schemes. These schemes may include batch, round robin, first in, first out, etc. In some embodiments, the priority scheme is based on meeting regulatory requirements. In some examples, the priority scheme includes rules based on financial entitlement symbols, order size, price, priority, client, pricing schemes, contractual obligations, etc.
[0317] In some embodiments, the selection of a wave from a queue may be based on the total routing / handling time for a particular wave. For example, if a wave has narrow timing parameters (i.e., small delays between data processing segments, and therefore shorter handling times), it may be prioritized over waves with longer handling times if it can easily fit into an upcoming schedule or result in higher throughput.
[0318] At 850, the system routes each of the data processing segments in the ordered data processing wave based on the schedule.
[0319] As described herein, the processor monitors data associated with the networked computing resource. In some embodiments, the processor derives the minimum handling interval time by identifying when the monitored data results in one or more latencies of two or more data processing segments routed to the same networked computing resource that are greater than a historical latency value for the networked computing resource. The difference in time between the related data processing segments can be used to define the minimum handling time for the networked computing resource. In some examples, the greater latency may indicate that one of the data processing segments has been queued at the networked computing resource (or along a path).
[0320] Similarly, in one embodiment, the processor may identify the number of data processing segments that can normally be routed to the same networked computing resource within a minimum handling time before longer than normal latencies are observed, which in one example may indicate, for example, the number of processors at the networked computing resource that should be busy before subsequent requests are queued.
[0321] In one embodiment, when monitored data (e.g., latency, acquisition ratio) indicates that a networked computing resource (or route) is congested, the processor may delay, reschedule, or re-create any wave that has at least one data processing segment for routing to the affected networked computing resource.
[0322] In some embodiments, monitoring the data includes identifying orders in various data processing segments from scheduled waves that should be routed to the same destination or have the potential to act as good test cases for determining timing parameters, handling times, congestion thresholds, etc. In some examples, by identifying, and specifically monitoring, these data processing segments with test capabilities, the system may in some cases reduce the need to send test data processing segments.
[0323] Although the present disclosure has been provided and illustrated in connection with certain currently preferred embodiments, many variations and modifications may be made without departing from the spirit and scope of the invention disclosed herein. The present disclosure and invention(s) are therefore not limited to the exact components or details of the methodology or configuration described above. No particular order to the steps or stages of the methods or processes described in this disclosure, including the drawings, is intended or implied, other than to the extent necessary or inherent in the process itself. In many cases, the order of process steps can be changed without changing the purpose, effect, or significance of the methods described. The scope of the claims should be defined solely by the appended claims in full consideration of the doctrine of equivalents and related theories.< / sell> < / sell> < / sell> < / sell> < / sell> < / sell> < / sell> < / buy> < / sell> < / buy> < / sell> < / sell> < / sell> < / buy> < / buy> < / buy> < / buy> < / sell> < / sell> < / sell> < / sell> < / sell> < / buy> < / buy> < / sell> < / buy> < / sell> < / buy> < / sell> < / buy> < / sell> < / sell> < / buy> < / sell> < / sell> < / buy> < / sell>
Claims
1. 1. A system for coordinating processing of data by multiple networked computing resources, the system comprising at least one processor, the processor comprising: obtaining a plurality of groups of data processing segments, each group including two or more data processing segments to be routed to one or more corresponding networked computing resources and a group timing sequence along which the two or more data processing segments are to be routed; obtaining a minimum handling interval for each of a plurality of networked computing resources; scheduling a multi-group timing sequence for routing two or more data processing segments included in each of the plurality of groups of data processing segments by determining a sequence for interspersing data processing segments of different groups without disturbing the group timing sequences or minimum handling intervals of the plurality of networked computing resources; routing each of the data processing segments in the multiple groups of data processing segments according to a scheduled multi-group timing sequence; A system configured to:
2. 2. The system of claim 1, wherein the group timing sequence identifies one or more time ranges during which a first data processing segment in a particular group of data processing segments can be routed relative to a time during which a second data processing segment in the particular group is routed.
3. Obtaining the plurality of groups of data-processing segments includes obtaining M groups of data-processing segments from a group queue; Scheduling the multi-group timing sequence for routing the M groups comprises: determining a total handling time for each possible arrangement of the M groups; scheduling the multi-group timing sequence as an arrangement among the possible arrangements having a shortest total handling time; The system of claim 1 , comprising:
4. The system of claim 3 , wherein M is selected based on a scheduling calculation time and a total handling time for one or more of the plurality of groups.
5. Obtaining the plurality of groups of data-processing segments includes obtaining M groups of data-processing segments from a group queue; Scheduling the multi-group timing sequence for routing the M groups comprises: determining a total handling time for the different arrangements of the M groups before a defined computation time has elapsed; scheduling the multi-group timing sequence as an arrangement from the different arrangements having a determined shortest total handling time before the defined computation time expires; The system of claim 1 , comprising:
6. The at least one processor monitoring data associated with the plurality of networked computing resources, the monitored data including data associated with data processing segments previously routed to the plurality of networked computing resources; The system of claim 1 configured to:
7. The at least one processor is configured to obtain a minimum handling interval time, and obtaining the minimum handling interval time includes: determining when two or more data processing segments routed to one of the plurality of networked computing resources resulted in one or more execution latencies greater than a historical latency value; defining a minimum handling time for one of the plurality of networked computing resources using the relative timing of the routing of the two or more data processing segments; The system of claim 6 , comprising:
8. The at least one processor identifying two or more data processing segments scheduled to be routed to one of the plurality of networked computing resources with relative timing that is within a threshold of a minimum handling interval of one of the plurality of networked computing resources; monitoring execution latencies for the two or more data processing segments to determine whether to adjust a minimum handling interval of one of the plurality of networked computing resources after routing the two or more data processing segments; The system of claim 1 configured to:
9. The at least one processor identifying a plurality of data processing segments scheduled to be routed to one of the plurality of networked computing resources with relative timing that is within a threshold of a minimum handling interval of one of the plurality of networked computing resources; monitoring execution latencies for the plurality of data processing segments to determine a number of data processing segments that will trigger a minimum handling interval of one of the plurality of networked computing resources after routing the plurality of data processing segments; The system of claim 1 configured to:
10. 2. The system of claim 1, wherein the multiple groups of data processing segments are obtained from a group queue, and the at least one processor is configured to fill the group queue with at most N groups of data processing segments from each of a plurality of session queues associated with one or more corresponding data sources.
11. 11. The system of claim 10, wherein filling the group queues includes selecting groups of data processing segments from the plurality of session queues using a priority scheme based at least in part on timing parameters associated with the groups.
12. 1. A method for coordinating processing of data by multiple networked computing resources, the method comprising: obtaining a plurality of groups of data processing segments, each group including two or more data processing segments to be routed to one or more corresponding networked computing resources and a group timing sequence along which the two or more data processing segments are to be routed; obtaining a minimum handling interval for each of a plurality of networked computing resources; scheduling a multi-group timing sequence for routing two or more data processing segments included in each of the plurality of groups of data processing segments by determining a sequence for interspersing data processing segments of different groups without disturbing the group timing sequences or minimum handling intervals of the plurality of networked computing resources; routing each of the data processing segments in the multiple groups of data processing segments according to a scheduled multi-group timing sequence; A method comprising:
13. 13. The method of claim 12, wherein the group timing sequence identifies one or more time ranges during which a first data processing segment in a particular group of data processing segments can be routed relative to a time during which a second data processing segment in the particular group is routed.
14. Obtaining the plurality of groups of data-processing segments includes obtaining M groups of data-processing segments from a group queue; Scheduling the multi-group timing sequence for routing the M groups comprises: determining a total handling time for each possible arrangement of the M groups; scheduling the multi-group timing sequence as an arrangement among the possible arrangements having a shortest total handling time; The method of claim 12, comprising:
15. 15. The method of claim 14, wherein M is selected based on a scheduling computation time and a total handling time for one or more of the plurality of groups of data processing segments.
16. Obtaining the plurality of groups of data-processing segments includes obtaining M groups of data-processing segments from a group queue; Scheduling the Group Timing Sequence for routing the M groups comprises: determining a total handling time for the different arrangements of the M groups before a defined computation time has elapsed; scheduling the group timing sequence as an arrangement from the different arrangements having a determined shortest total handling time before the defined computation time expires; The method of claim 12, comprising:
17. monitoring data associated with the plurality of networked computing resources, the monitored data including data associated with data processing segments previously routed to the plurality of networked computing resources; The method of claim 12, comprising:
18. The method includes obtaining a minimum handling interval time, the obtaining the minimum handling interval time comprising: determining when two or more data processing segments routed to one of the plurality of networked computing resources resulted in one or more execution latencies greater than a historical latency value; defining a minimum handling time for one of the plurality of networked computing resources using the relative timing of the routing of the two or more data processing segments; 20. The method of claim 17, comprising:
19. identifying two or more data processing segments scheduled to be routed to one of the plurality of networked computing resources with relative timing that is within a threshold of a minimum handling interval of one of the plurality of networked computing resources; monitoring execution latencies for the two or more data processing segments to determine whether to adjust a minimum handling interval of one of the plurality of networked computing resources after routing the two or more data processing segments; The method of claim 12, comprising:
20. identifying a plurality of data processing segments scheduled to be routed to one of the plurality of networked computing resources with relative timing that is within a threshold of a minimum handling interval of one of the plurality of networked computing resources; monitoring execution latencies for the plurality of data processing segments to determine a number of data processing segments that will trigger a minimum handling interval of one of the plurality of networked computing resources after routing the plurality of data processing segments; The method of claim 12, comprising:
21. 13. The method of claim 12, wherein the multiple groups of data processing segments are obtained from a group queue, the group queue being filled with at most N groups of data processing segments from each of a multiple session queues associated with one or more corresponding data sources.
22. 22. The method of claim 21, wherein filling the group queues comprises selecting groups of data processing segments from the plurality of session queues using a priority scheme based at least in part on timing parameters associated with the groups.
23. One or more non-transitory computer-readable media having instructions stored thereon that, when executed by at least one processor, cause the at least one processor to: obtaining a plurality of groups of data processing segments, each group including two or more data processing segments to be routed to one or more corresponding networked computing resources and a group timing sequence along which the two or more data processing segments are to be routed; obtaining a minimum handling interval for each of the networked computing resources; scheduling a multi-group timing sequence for routing two or more data processing segments included in each group of a plurality of groups of the plurality of data processing segments by determining a sequence for interspersing the data processing segments of different groups without disturbing the group timing sequence or a minimum handling interval of the plurality of networked computing resources; routing each of the data processing segments in the multiple groups of data processing segments according to a scheduled multi-group timing sequence; A computer-readable medium configured to:
Citation Information
Patent Citations
Data synchronization using networked computing resources
JP2013513171A
